UNIVERSIDADE FEDERAL DE SANTA CATARINA PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA
DE AUTOMAÇÃO E SISTEMAS
Amadeu Plácido Neto
AN INTERNAL LEAKAGE FAILURE
PROGNOSIS SYSTEM FOR
ELECTRO-HYDROSTATIC SUBSEA
VALVE ACTUATORS
Florianópolis 2018
Amadeu Plácido Neto
AN INTERNAL LEAKAGE FAILURE
PROGNOSIS SYSTEM FOR
ELECTRO-HYDROSTATIC SUBSEA VALVE
ACTUATORS
Dissertação submetida ao Pro-grama de Pós-Graduação em En-genharia de Automação e Sistemas da Universidade Federal de Santa Catarina para a obtemção do grau de Mestre em Engenharia de Au-tomação e Sistemas.
Orientador: Prof. Ubirajara Franco Moreno, Dr. Coorientador: Alexandre Orth, Dr.
Florianópolis 2018
Ficha de identificação da obra elaborada pelo autor,
através do Programa de Geração Automática da Biblioteca Universitária da UFSC.
Plácido Neto, Amadeu
An internal leakage failure prognosis system for electro-hydrostatic subsea valve actuators / Amadeu Plácido Neto ; orientador, Ubirajara Franco Moreno, coorientador, Alexandre Orth, 2018.
118 p.
Dissertação (mestrado) - Universidade Federal de Santa Catarina, Centro Tecnológico, Programa de Pós Graduação em Engenharia de Automação e Sistemas, Florianópolis, 2018.
Inclui referências.
1. Engenharia de Automação e Sistemas. 2. Prognóstico de Falha. 3. Sistema de Produção Submarino. 4. Manutenção Baseada em Condição. I. Franco Moreno, Ubirajara. II. Orth, Alexandre. III. Universidade Federal de Santa Catarina. Programa de Pós-Graduação em Engenharia de Automação e Sistemas. IV. Título.
Amadeu Plácido Neto
AN INTERNAL LEAKAGE FAILURE PROGNOSIS SYSTEM FOR ELECTRO-HYDROSTATIC SUBSEA
VALVE ACTUATORS
Esta Dissertação foi julgada adequada para a obtenção do Título de "Mestre em Engenharia de Automação e Sistemas" e aprovada em sua forma final pelo Programa de Pós-Graduação em Engenharia de Au-tomação e Sistemas.
Florianópolis, 03 de Outubro de 2018.
Prof. Dr. Werner Kraus Junior
Coordenador do Programa de Pós-Graduação
Banca examinadora:
Prof. Ubirajara Franco Moreno, Dr.
Orientador
Prof. Felipe Gomes de Oliveira Cabral, Dr.
Universidade Federal de Santa Catarina
Prof. Marcelo Ricardo Stemmer, Dr.
Universidade Federal de Santa Catarina
Prof. Victor Juliano de Negri, Dr.
AGRADECIMENTOS
Eu dedico esta dissertação em primeiro lugar aos meus pais, Amadeu Plácido Filho e Márcia Gonçalves Plácido e às minhas irmãs, Juliana Gonçalves Plácido e Carolina Gonçalves Plácido. Agradeço sincera-mente pelo apoio integral e virtualsincera-mente incondicional que sempre me foi provido.
Agradeço aos meus colegas de trabalho – na empresa e na univer-sidade – sobretudo ao meu coorientador Alexandre Orth e meu ori-entador Ubirajara Franco Moreno, sempre muito pacientes, deram-me a oportunidade e depositaram a confiança em mim para concluir este estudo.
Sou grato por fim aos meus amigos e amigas – Henrique, Bruno L., Eduardo, Bruno G., Rafael, Bárbara, Rodrigo, Marcelo e inúmeros outros – que partilharam comigo todo o espectro de sentimentos, da angústia até a alegria, típicos durante um trabalho de mestrado.
Todos eles e elas merecem nada menos do que o melhor de mim. Eu não acredito que seja o que apresento aqui – 2018 me foi particularmente insólito, quase como um rito de passagem – todavia, saibam que foi o melhor que pude fazer nas minhas circunstâncias.
"He, that will not sail till all dangers are over, must never put to sea."
RESUMO
Os custos de manutenção e reparo de equipamentos utilizados na indústria de petróleo & gás offshore – particularmente em sistemas de produção submarinos (SPS) – correspondem a um gasto significativo do custo operacional (OPEX). Este fator motiva a demanda de equipa-mentos que possibilitam estratégias de manutenção melhoradas, como uma estratégia de manutenção baseada em condição (CBM). CBM tem como objetivo avaliar a condição do equipamento e prever o seu tempo de vida útil remanescente (RUL) para agendar um procedimento de manutenção em tempo hábil. Ela pode ser implementada através de temas de monitoramento de condição, que por sua vez fazem uso de sis-temas de diagnóstico de avarias e sissis-temas de prognóstico de falhas. Os atuadores de válvula abrangem partes importantes da infra-estrutura de SPSs. Foi observado que vazamento hidráulico é o mecanismo de falha mais comum acometido sobre este equipamento. Nesta pesquisa, um atuador eletro-hidrostático de válvulas submarinas é analisado e um sistema de prognóstico de vazamento interno no seu cilindro hidráulico é proposto. O sistema de prognóstico é modelado de acordo com testes de degradação acelerada da vedação do pistão do atuador – onde acredita-se que acredita-seja a fonte da falha – e de acordo com o modelo fenomenológico do atuador. O sistema é implementado e validado através de múltiplas simulações aleatórias numa abordagem Monte Carlo. Os resultados para a melhor configuração do sistema reportaram um RUL confiável em 85% dos experimentos – 85 de 100 experimentos – com uma ante-cedência média de 18,45 meses e um erro médio de 2,01 meses antes do verdadeiro evento de falha.
Palavras-chave: Prognóstico de Falha, Identificação de Sistemas,
Manutenção Baseada em Condição, Sistema de Produção Submarino, Atuadores Eletro-Hidrostáticos.
RESUMO ESTENDIDO
Introdução
Os custos de manutenção e reparo de equipamentos utilizados na in-dústria de petróleo & gás offshore – particularmente em sistemas de produção submarinos (SPS) – correspondem a um gasto significativo do custo operacional (OPEX). Este fator motiva a demanda de equipa-mentos que possibilitam estratégias de manutenção melhoradas, como uma estratégia de manutenção baseada em condição (CBM). CBM tem como objetivo avaliar a condição do equipamento e prever o seu tempo de vida útil remanescente (RUL) para agendar um procedimento de manutenção em tempo hábil. Ela pode ser implementada através de sistemas de monitoramento de condição, que por sua vez fazem uso de sistemas de diagnóstico de avarias e sistemas de prognóstico de fal-has. Os atuadores de válvula abrangem partes importantes da infra-estrutura de SPSs.
Objetivos
O objetivo geral deste trabalho é elaborar e avaliar um esquema de prognóstico de falha de vazamento interno em um atuador de válvula submarino de princípio eletro-hidrostático utilizando um algoritmo de estimação de parâmetro de caixa-cinza e uma regressão linear para a predição da vida útil restante.
Metodologia
A metodologia empregada na execução do trabalho – i.e., na elabo-ração do sistema de prognóstico – é dividida em duas partes: projeto & implementação e verificação & validação. A primeira trata-se ini-cialmente de um estudo sobre o processo, sua distribuição de falhas e seus mecanismos de degradação. Com este conhecimento heurístico, são avaliados os algoritmos e técnicas adequadas para o prognóstico em si, e eventualmente a modelagem formal do sistema. A segunda
viii
parte foca na definição de métricas e procedimentos para verificar se o sistema concebido cumpre seus requisitos e avaliar o seu desempenho. Com os resultados obtidos, as duas partes da metodologia podem ser novamente iteradas com o objetivo de aprimorar o desempenho do sis-tema.
Resultados e discussão
Uma vez estudado o processo e suas falhas, o mecanismo de degradação das vedações internas pode ser testado em laboratório. Foram elabora-dos modelos matemáticos do processo e do mecanismo de degradação e estabelecida premissas sobre a relação entre a degradação da vedação interna e o efeito na dinâmica do processo. Foram definidos os requi-sitos do sistema de prognóstico de falha e a estrutura do algoritmo, a partir dos quais foram definidas e sintonizadas as técnicas de estimação de parâmetro (estimação em caixa-cinza por método de erro-predição) e de predição de vida útil remanescente (regressão linear por mínimos quadrados ordinários). Finalmente na etapa de verificação e validação foram apresentadas as métricas para avaliar o modelo do processo e o desempenho do sistema de prognóstico para diferentes configurações. A verificação do sistema de prognóstico foi avaliada através de uma abor-dagem Monte Carlo, com sucessivas simulações de cenários aleatórios. O sistema de prognóstico foi avaliado para três janelas de amostragem, vinte e quatro, trinta e seis, e quarenta e oito meses. A menor janela de amostragem apresentou uma porcentagem de convergência maior do que as demais, embora o erro médio de predição final tenha sido mais do que três vezes maior do que o obtido com a maior janela de amostragem. Esta configuração também apresentou a menor porcen-tagem de convergência em tempo hábil. De um modo geral, a janela de amostragem de trinta e seis meses apresentou o melhor resultado entre todas as métricas de avaliação de desempenho.
Considerações finais
Todos os objetivos propostos foram cumpridos rigorosamente. O resul-tado obtido, porém, está atrelado a uma série de hipóteses e premissas realizadas durante a etapa de projeto e implementação. Tais
premis-ix
sas devem ser investigadas como trabalho futuro para estabelecer um escopo de validade mais bem-definido para o sistema implementado.
Ademais, outros métodos e algoritmos de estimação de parâmetros e predição da vida útil remanescente podem ser estudados com o intu-ito de eliminar a necessidade das premissas utilizadas e refinar ainda mais os resultados obtidos.
Palavras-chave: Prognóstico de Falha, Identificação de Sistemas,
Manutenção Baseada em Condição, Sistema de Produção Submarino, Atuadores Eletro-Hidrostáticos.
ABSTRACT
In the Offshore Oil & Gas industry, particularly for Subsea Produc-tion Systems (SPS), the cost of maintenance and repair of equipment corresponds to a significant amount of the Operational Expenditure (OPEX). This drives the demand for equipment that enable improved maintenance strategies, such as a Condition-Based Maintenance (CBM) strategy. CBM aims to assess the actual condition of the equipment and forecast its remaining useful life (RUL) in order to schedule a main-tenance intervention in a timely manner. It can be achieved with con-dition monitoring, which employs both fault diagnosis systems as well as failure prognosis systems. The valve actuators are important pieces of equipment in SPSs. It has been observed that leakage is the most common failure mechanism inflicted on this kind of equipment. In this research, an electro-hydrostatic subsea valve actuator is analyzed and a failure prognosis framework for internal leakage in the hydraulic cylin-der is proposed. The prognosis system is modeled after Accelerated Degradation Tests of the piston sealing, which is believed to be the source of the failure, and the phenomenological description of the ac-tuator. The system is implemented and validated through randomized simulation in a Monte Carlo approach. The results for the best config-uration reported a reliable RUL estimation in 85% of the experiments – 85 out of 100 experiments performed – with an average of 18.45 months of antecedence and an average error of 2.01 months prior to the failure event.
Keywords: Failure Prognosis, System Identification,
Condition-Based Maintenance, Subsea Production System, Electro-Hydrostatic Actuator.
Contents
1 Introduction 1
1.1 Motivation . . . 1
1.2 Maintenance Strategies . . . 2
1.3 Failure Prognosis Systems . . . 5
1.4 State-of-the-art on Internal Leakage Prognosis of Hydraulic Cylinders . . . 7
1.5 Objectives & Contributions . . . 10
1.6 Methodology . . . 10
1.7 Structure . . . 12
2 Subsea Valve Actuator and the Prognosis System Framework 15 2.1 Subsea Valve Actuators . . . 15
2.1.1 Failure Distribution of Subsea Valve Actuators . . . 18
2.1.2 Piston Sealing Degradation . . . 20
2.2 Prognosis System Framework . . . 25
2.3 Mathematical Models . . . 27
2.3.1 Degradation Model . . . 28
2.3.2 Process Model . . . 30
3 Implementation of the Internal Leakage Prognosis System 41 3.1 System Requirements . . . 41
xiv CONTENTS
3.2 General Algorithm . . . 42
3.3 Parameter Identification Method . . . 44
3.3.1 Gray-Box Model . . . 45 3.3.2 Excitation Signal . . . 46 3.3.3 Estimation Uncertainty . . . 48 3.4 Forecast Method . . . 49 3.4.1 Sampling Window . . . 51 3.4.2 Prediction Interval . . . 52
4 Verification & Validation 59 4.1 Process Model Validation . . . 59
4.2 Prognosis System Verification . . . 64
4.2.1 Test Scenario . . . 64
4.2.2 Performance Metrics . . . 67
4.2.3 Results & Discussion . . . 68
5 Conclusion 77 5.1 Summary . . . 77
List of Figures
1.1 Influence of downtime due to failure in the profit return of a process/plant (adapted from [1]). . . 3 1.2 The OSA-CBM structure for Condition Monitoring
solu-tions [2]. . . 9 1.3 A systematic approach to implement a generic failure
prog-nosis system (loosely based on [3]). . . 11 2.1 The patented concept of the electro-hydrostatic subsea valve
actuator [4]. . . 16 2.2 The simplified electro-hydrostatic valve actuator circuit. . . 17 2.3 The test bench for verification and validation of the novel
Subsea Valve Actuator (SVA) design. . . 18 2.4 The failure mechanism distribution for components
consti-tuting SVAs [5]. . . 19 2.5 Sealing effect of an inactive (on the left) and active (on the
right) O-ring (adapted from [6]). . . 21 2.6 The degradation profile of the elastic modulus of an O-ring
immersed in different fluids. . . 24 2.7 An illustration of the information flow for the proposed
fail-ure prognostic framework. . . 27 xv
xvi LIST OF FIGURES
2.8 A family of 15 randomly generated degradation profiles computed from equation (2.7). . . 31 2.9 The process circuit with the gate valve (not in actual
pro-portion), highlighted control volumes and state designators. 33 3.1 The algorithmic representation of the internal leakage
prog-nosis system. . . 43 3.2 The gray-box hydraulic circuit model. . . 46 3.3 The Partial Stroke Test (PST) control loop architecture. . 47 3.4 The algorithmic representation of the PST for position
con-trol. . . 47 3.5 A series of estimated parameters in contrast with the actual
parameter value. . . 49 3.6 The linear regression prediction for the degradation process. 50 3.7 A case where a shorter window has a better approximation
of the future values. . . 53 3.8 A case where a larger window has a better approximation
of the future values. . . 54 3.9 Histograms and normal distribution fit of the error between
the regression prediction and the samples for two sampling window sizes. . . 55 3.10 A randomized simulation with the Remaining Useful Life
(RUL) estimations and their prediction intervals with a sampling window of 48 samples. . . 57 4.1 Comparison between the phenomenological model and the
physical test bench with NRMSE values. . . 61 4.2 Comparison between the phenomenological model and the
physical test bench with NRMSE values. . . 62 4.3 Comparison between the phenomenological model and the
physical test bench with NRMSE values. . . 63 4.4 The information flow of the prognosis framework with the
performance evaluation blocks. . . 65 4.5 The algorithmic representation of the internal leakage
prog-nosis system. . . 66 4.6 Convergence rate of experiments. . . 69 4.7 Box-whisker plot of the RUL estimation error. . . 70
LIST OF FIGURES xvii
4.8 The most inaccurate experiment with 24 samples window forecast. . . 72 4.9 The most inaccurate experiment with 48 samples window
forecast. . . 73 4.10 Box-whisker plot of the prediction interval width. . . 74 4.11 Timely estimation proportion from the total converged
List of Tables
2.1 Definitions of the failure mechanism designators according to Annex B of [7]. . . 20 3.1 Relative errors for the regression model predictions at
dif-ferent time intervals in Figure 3.6. . . 51 4.1 The FPR and FNR values. . . 71 4.2 The summarized performance score for all tested forecast
window sizes. . . 72
Acronyms
ADT Accelerated Degradation Test. CAPEX Capital Expenditure.
CBM Condition-Based Maintenance. EHA Electro-Hydrostatic Actuator. FNR False Negative Ratio.
FPR False Positive Ratio.
IMR Intervention, Maintenance and Repair. OLS Ordinary Least Squares.
OPEX Operational Expenditure. PEM Prediction-Error Minimization. PHM Prognostics & Health Management. PoF Physics-of-Failure.
PST Partial Stroke Test.
RNG Random Number Generator. RUL Remaining Useful Life. SPS Subsea Production Systems. SVA Subsea Valve Actuator.
xxii Acronyms
TTF Time To Fail. UUT Unit Under Test.
Symbols
H0 Alternative hypothesis of the statistical test. H0 Null hypothesis of the statistical test.
K Number of samples obtained from the physical test bench. Tq Student’s t-test score for prediction bounds.
¯y Arithmetic mean value of the physical test bench output over a num-ber of samples K.
β0 Intercept parameter of linear regression. η Significance level of prediction bounds. G0 Reaction kinetics-related parameter. R Reaction rate according to Arrhenius law. AC System parameter – cylinder piston area.
CM System parameter – motor driver filter capacitance. E Activation energy factor.
JM System parameter – motor. xxiii
xxiv List of symbols
LM System parameter – motor windings inductance. R2 Coefficient of determination.
R100 % Cylinder leakage resistance measured at 100% of the nominal
elastic modulus of the sealing on nominal operation conditions. R60 % Cylinder leakage resistance at 60% of the nominal elastic
mod-ulus of the sealing on nominal operation conditions. RCL System parameter – cylinder leakage resistance. RM System parameter – motor windings resistance. R Universal gas constant.
Ta Accelerated degradation test temperature. Tr Nominal degradation temperature.
TM System parameter – motor torque constant. W Load torque in the motor shaft.
∆ˆp Measured differential pressure in the cylinder chambers. ∆p State variable – difference between pTL and pTR. Γmax Maximum random degradation increase.
Γ Discrete random variable – degradation rate increase. ΩM System parameter – motor velocity constant. Φ Discrete random variable – cylinder load disturbance. Υmax Maximum random degradation rate uncertainty. Υmin Minimum random degradation rate uncertainty. Υ Discrete random variable – degradation rate uncertainty. αP System parameter – motor-pump transmission ratio. ¯aV Area of the orifice of the piloted check valve.
List of symbols xxv
χ RUL estimation uncertainty.
δp Additive Gaussian noise in the manometer output. δq Additive Gaussian noise in the flowmeter output.
δv Additive Gaussian noise in the translational velocity output of the encoder.
δx Additive Gaussian noise in the position output of the encoder. δ Measurement noise vector.
C System parameter – tuning parameter for the cylinder cushion model. γ Discrete random variable – degradation rate.
ˆ
θ Estimated process parameter vector. ˆ
f Gray-box model of the process for Prediction-Error Minimization algorithms.
ˆ
q Measured flow in the cylinder inlet port. ˆ
q Measured position of the cylinder rod. ˆ
q Measured translational velocity of the cylinder rod. ˆtRUL Estimated RUL.
ˆt Estimated parameter degradation time vector. ˆ
xC System parameter – maximum cylinder stroke length. λ Elastic modulus.
UΓ Probability density function of Γ. UΥ Probability density function of Υ.
ω State variable – rotational frequency of motor shaft. φ Nominal load on the cylinder rod.
ψ Vectors of parameters to be estimated in Prediction-Error Minimiza-tion algorithms.
xxvi List of symbols
ρ System parameter – system hydraulic fluid density.
τp Cutoff frequency of low-pass filter in the manometer output. τq Cutoff frequency of low-pass filter in the flowmeter output.
τv Cutoff frequency of low-pass filter in the translational velocity out-put of the encoder.
τx Cutoff frequency of low-pass filter in the position output of the en-coder.
θ(t) Degradation percentage of the O-Ring seal.
θmax Maximal tolerated deviation of the parameter, beyond which con-figures a faulty state.
{Ψmax, Ψmin} Vectors of parameter constraints for Prediction-Error Minimization algorithms.
cD System parameter – piloted check valve discharge coefficient. dP System parameter – pump viscous drag.
fC System parameter – cylinder friction coefficient. fM System parameter – motor-pump viscous drag.
gV System parameter – tuning parameter for the piloted check valve. i State variable – electrical current in the motor windings.
kV System parameter – tuning parameter for the piloted check valve. k Discrete time variable denoting higher-frequency sampled signals. mC System parameter – cylinder piston mass.
n Discrete time variable denoting monthly sampled signals. pTL State variable – top-left pressure.
pTR State variable – top-right pressure.
List of symbols xxvii
qC Flow rate through the cylinder chambers. qPL Internal leakage flow rate of the pump. qP Flow rate output of the pump.
rPST Partial stroke test set-point. rv Velocity controller set-point. rx Position controller set-point. r Process set-point vector.
ta Accelerated degradation test time. tr Nominal degradation time. u Process input vector.
v State variable – cylinder piston velocity. w Process load disturbance.
y Process output vector.
z State variable – input commutation electrical voltage. σ Standard deviation.
˜y Output of the physical test bench.
mf Final index of regression sampling window. mi Initial index of regression sampling window.
CHAPTER
1
Introduction
1.1
Motivation
Recent reports estimate that, by 2030, 48% of all oil production and 42% of all natural gas production will occur offshore. The same reports forecast a global growth of energy demand of 40%, where coal, oil and natural gas still figure as the main energy sources [8]. These trends drive the development of increasingly more efficient production of these resources. As part of these efforts, the development of technologies to reduce maintenance costs of Subsea Production Systems (SPS) is a pri-mary objective. Moreno-Trejo et al. [9] highlight the huge potential of improvement of Intervention, Maintenance and Repair (IMR) strate-gies in SPSs. The authors explicit the benefits of preventive mainte-nance and Condition-Based Maintemainte-nance (CBM) strategies in contrast to conventional corrective maintenance strategy to reduce downtime and economically improve the Oil & Gas production in SPSs.
This research is part of an effort to elaborate and evaluate a method that can enable a Condition-Based Maintenance approach for the most commonly observed fault in drive components constituting a SPS in-frastructure. A particular electro-hydrostatic actuator design is used
2 CHAPTER 1. INTRODUCTION
as a case-study. This chapter gives a brief overview of maintenance strategies, failure prognosis systems and their terminology in general, a background of the state-of-the-art in prognosis of internal leakage of hydraulic cylinders – which is the object of the case study – and fin-ishes with the objectives and contributions of the research as well as a description of the structure of the document.
1.2
Maintenance Strategies
Maintenance (also maintenance strategy or maintenance policy) is un-derstood within the scope of this text as defined by [7]:
"The combination of all technical and management ac-tions intended to retain an item in, or restore it to, a state in which it can perform as required."
Maintenance strategies comprise a field of particular interest to the industry, as an improvement in the maintenance policy of an asset is directly related to the growth of its profit. This is illustrated in terms of Capital Expenditure (CAPEX) – the cost for the purchase of an asset – and Operational Expenditure (OPEX) – the cost to operate this asset. In Figure 1.1 it is shown the ideal scenario on the superior side, where a process/plant never fails, versus a more realistic scenario on the inferior side, where a failure event occurs. The downtime generated by the failure event and by the maintenance procedure greatly delays the break-even of the process/plant.
In order to discuss objectively the maintenance strategies, a ter-minology must be introduced to reassure consistency and avoid mis-interpretation. The following definitions of Fault and Failure are also reproduced from ISO 14224 [7]1:
• Failure – The loss of ability to perform as required. A failure of an item is an event that results in a fault of that item.
• Fault – The inability to perform as required, due to an internal state. A fault of an item results from a failure, either of the item itself, or from a deficiency in an earlier stage of the life cycle, such as specification, design, manufacture or maintenance.
1.2. MAINTENANCE STRATEGIES 3
Figure 1.1: Influence of downtime due to failure in the profit return of a process/plant (adapted from [1]).
Moreover, a Fault may develop abruptly (step-wise) or incipiently (drift-wise). This classification of a fault’s dynamic must be given in respect to its detection system limitations (ex.: sample rate) and the time required to react (ex.: maintenance scheduling).
However, there is not a consensus over this nomenclature. It is not rarely seen in the literature the almost exact definition of both terms given the other way around. This is the case in the standard ISO/IEC/IEEE 24765 [11], which curiously, in contradiction with ISO 14224, defines a fault as an event that precedes a state of failure. It must be considered that, although these standards are issued by the same institution, the issuing technical committees and their area of expertise differs between themselves. What is common in the present
4 CHAPTER 1. INTRODUCTION
literature is the causal relationship of an event that leads to a state in which a system cannot perform its functionalities as required. In this document the nomenclature defined above according to ISO 14224 shall be used, as it is more specifically devised for the Oil & Gas industry.
In addition, Watton [1] classifies three main maintenance strategies which can be applied to whichever general industrial setting: correc-tive maintenance, prevencorrec-tive maintenance and condition-based main-tenance.
Corrective maintenance is the most conventional maintenance pol-icy. It consists in allowing the system to continue operating until a failure occurs. In this occasion, repairs or substitutions are executed as necessary. The drawbacks in contrast to the simplicity of imple-mentation of this approach are the long-term unpredictability of the downtimes. Failures tend to occur on inconvenient moments (such as system under-stress due to high demand). Depending on the situation, this can lead to idle manpower or bottlenecks in the production line. Furthermore, components, tools and specialized personnel required to fulfill such repairs and substitutions must be either available at all times to reduce the downtime, increasing the costs associated to stor-age, idle assets and staff, or else they can be ordered on demand, which leaves the maintenance process at the mercy of the availability of such resources, delaying even more the production downtime. Finally, un-expected failures may lead to catastrophic events, which increase the risk of incidents in the workspace.
Preventive maintenance is a policy established under a strict main-tenance schedule, during which each assigned component is replaced/re-paired in predetermined time intervals. Such intervals are defined through a combination of manufacturer’s data and statistical and his-torical operation data derived from a population of the same component under similar operational circumstances. With a planned maintenance, it is possible to control the interruption time in a plant and prepare it so that the maintenance is carried out swiftly. The necessary components for replacement can be acquired shortly beforehand, reducing the ne-cessity of storage and the uncertainty of their availability. There is also the possibility to hire a specialized third-party crew to carry out the maintenance service. This method still may encompass a potentially high cost when it replaces equipment that are still in good conditions.
1.3. FAILURE PROGNOSIS SYSTEMS 5
A functional component can be refurbished to restore it to its origi-nal characteristics, however the predictability of its service life, which is used to determine the replacement schedule, may be compromised. There is also the possibility of a component failure before the scheduled maintenance, which draws back to the corrective maintenance proce-dure. These occurrences are, however, significantly rarer than when no preventive maintenance strategy is implemented.
Finally, the CBM strategy (also known in the literature as Pre-dictive Maintenance) incorporates monitoring technologies to estimate with a higher accuracy the remaining service life of a component. With this information, the components can be maintained with an adaptive schedule, preventing the replacement of items that are still in perfectly – or even marginally – good condition. The data may also allow the plant operator to adapt the set points of the system in order to stretch the service life, in case there is no replacement part/crew available in the short-term future. CBM is effectively achieved by implementing a Condition Monitoring System, which comprises Fault Diagnosis Sys-tems and Failure Prognosis SysSys-tems. The first is a more established technology, particularly effective for CBM of fault-tolerant systems. The latter, however, has a lower degree of maturity, mostly because of the inherent difficulty to obtain a reliable prediction of stochastic pro-cesses [12]. The techniques utilized in the implementation of CBM are comprised in the Prognostics & Health Management (PHM) research field.
1.3
Failure Prognosis Systems
A concise definition for Failure Prognosis System within the PHM con-text can be quoted from [13]:
"Prognosis is the ability to predict accurately and pre-cisely the remaining useful life of a failing component or subsystem."
Component and subsystem are henceforth designated as Unit Under Test (UUT). The literature presents both terms Fault Prognosis and Failure Prognosisas semantically equivalent to each other. Throughout this text it is employed Failure Prognosis, as the life prediction concerns
6 CHAPTER 1. INTRODUCTION
to a time-related figure, and as defined before, "failure" is an event that occurs at a given time.
A Failure Prognosis System tracks the evolution of one or more features on an UUT and derives a quantitative figure that represent the time remaining from the current instant for a given failure event to occur in this UUT. A feature can be either a direct physical quantity – representing a crack length, clearance size or vibration amplitude for instance – or a virtual index representing the "health" of a subsystem or the entire UUT. The PHM terminology denominates this quantity as Remaining Useful Life (RUL) rather than Time To Fail (TTF), because the latter is already employed in the context of Reliability Engineering. Due to the stochastic nature of the degradation mechanisms, a RUL must be provided with an associated confidence interval.
According to Saxena [14], the requirements of a Failure Prognosis System can be defined by the following characteristics:
• Accuracy – the estimated RUL must be close to the actual time of the failure event;
• Precision – the confidence interval of the estimated RUL must be acceptably narrow;
• Robustness – the RUL estimation must be reliable and withstand disturbances and deviations from the nominal case within a given tolerance;
• Timeliness – the RUL estimation must be provided within an acceptable time-frame prior to the time of the failure event. The RUL is then fed to an advisory manager, which takes a decision on when to schedule a maintenance according to the estimation and its uncertainties.
There are generally two main classes of approaches to achieve a prac-tical implementation of a Failure Prognosis System: the Model-Based Approach and the Data-Driven Approach [15]. In short, the first makes use of empirical and analytical knowledge to create an explicit model of the UUT degradation over time, while the latter utilizes historical data and statistical inference to estimate the UUT degradation. These concepts are explained further in the document.
1.4. STATE-OF-THE-ART ON INTERNAL LEAKAGE PROGNOSIS OF
HYDRAULIC CYLINDERS 7
The Failure Prognosis System explored in this research is classified as model-based, mainly because the UUT treated is a novel equipment. The whole composition makes use of phenomenological modeling tech-niques, system identification techniques – particularly the non-linear gray-box system identification method – and simple regression analy-sis.
1.4
State-of-the-art on Internal Leakage Prognosis of
Hydraulic Cylinders
The internal leakage in hydraulic actuators has been traditionally treated in the scope of fault diagnosis (detection), a field regarded to have a relatively higher Technology Readiness Level [14]. That is, there are consolidated techniques which focus on the identification of a leakage after it has exceeded a threshold rather than predicting the time in which it will exceed it.
To identify and quantify the internal leakage in a hydraulic cylin-der, Sepehri et al. [16] utilized an Extended Kalman Filter scheme to compare the residuals in the pressure signals of the nominal model and the real process. Similarly, Vietor et al. [17] utilized spectral analy-sis of these residuals to detect the internal leakage in similar hydraulic cylinders. On another research, Sepehri et al. [18] utilized the Hilbert-Huang transform to identify an internal leakage feature pattern in the pressure signals of the hydraulic cylinder chamber. Later, Sepehri et al. [19] has employed time series cross-correlation on the chamber’s pressure signals to diagnose cross-port leakage. In particular for the purpose of internal leakage detection in Electro-Hydrostatic Actuator (EHA)s, Sepehri et al. [20] studied the possibility to use multi-scale analysis. Regarding the Oil & Gas industry, in his thesis, Stavenes [21] gives a holistic view of hydraulic leakage in the entire Subsea Control System of the SPS infrastructure, however the solutions proposed are also focused on the diagnosis rather than the prognosis.
In order to evaluate the state-of-the-art of the research in the prog-nosis of internal leakage and/or similar systems, a thorough biblio-graphic review has been performed on the most significant proceedings of the PHM community. In total 16 proceedings published between 2011 and 2017 were reviewed from the following symposia and
confer-8 CHAPTER 1. INTRODUCTION
ences:
• Annual Conference of the Prognostics and Health Management Society
• European Conference of the Prognostics and Health Management Society
• International Federation of Automatic Control Symposium on Fault Detection, Supervision and Safety for Technical Processes • Institute of Electrical and Electronic Engineers International
Con-ference on Prognostics and Health Management
• Condition Monitoring and Diagnostic Engineering Management Conference
The proceedings contained a total of 1274 published papers, from which a more broad filtering of keywords and titles yielded 21 papers related for this investigation. The contents of these papers were then assessed individually to assert their relevance. The searched keywords were regular expressions of "failure prognosis", "prognostic", "health man-agement", "electro-hydrostatic actuator", "hydraulic actuator" and "hy-draulic leakage".
Nearly all of the investigated studies are concentrated in the Aero-space industry. In general, the cycle demands, specifications and envi-ronment settings of a hydraulic system used in aircrafts are very dis-crepant from the Oil & Gas industry, thus both the failure distribution and the failure mechanisms are not comparable. For instance, the case of leakage prognosis of aircraft hydraulic systems has been addressed in the study of Vianna et al. [22]. The study considers the leakage in the whole hydraulic system instead of only cross-port leakage in the cylinders. The author states that the leakage is mainly caused by thermal stresses, which are not a main concern in the subsea Christ-mas trees. In the study of Mornacchi et al. [23], the author explains that most degradation processes in a typical aircraft electro-hydraulic servo-actuator tend to be shorting of electrical motor windings, con-tamination of hydraulic fluid, spring fatigue and wear of components in the solenoid control valve.
Still, all the techniques researched essentially fit in the classification given by Dragomir et al. [15], which can be summarized as:
1.4. STATE-OF-THE-ART ON INTERNAL LEAKAGE PROGNOSIS OF
HYDRAULIC CYLINDERS 9
• Model-Based Approach – Prognostic techniques in which there is a known model for the degradation process, acquired either analytically using Physics-of-Failure (PoF) or empirically with Accelerated Degradation Test (ADT). These approaches use sys-tem/parameter identification and extrapolation in order to pro-vide a RUL estimation.
• Data-Driven Approach – Prognostic techniques which are based in available field data of the degradation process. It is preferred when the system is too complex to either model analytically or identify the degradation mechanisms. These approaches typically use machine learning or statistical inference to provide a RUL estimation.
Furthermore, there are normative references to standardize the im-plementation of Condition Monitoring solutions – both Fault Diagnosis and Failure Prognosis Systems. The standards do not impose restric-tions regarding the technology (either data-driven or model-based ap-proaches), but aim to improve the interchangeability of the Condition Monitoring solutions.
Particularly, the standard ISO 13374-1 [2] defines structure of the condition monitoring solution as well as data flow and presentation between modules. It is based on the Open System Architecture for Condition-Based Maintenance (OSA CBM) and is structured as shown in the block diagram in Figure 1.2.
Data Acquisition (DA)
Data Manipulation (DM)
State Detection (SD)
Health Assessment (HA)
Prognostic Assessment (PA)
Advisory Generation (AG)
10 CHAPTER 1. INTRODUCTION
The Fault Diagnosis Systems are most commonly comprised within the Health Assessment layer, whereas the Failure Prognosis Systems lie in the Prognostic Assessment layer.
1.5
Objectives & Contributions
The general objective of this work is to elaborate and evaluate a frame-work for the prognostic of internal leakage in the cylinder of an electro-hydrostatic subsea valve actuator due to piston sealing degradation. The method to be implemented uses a gray-box parameter estimation algorithm cascaded with an ordinary linear regression for RUL predic-tion. In order to achieve this goal, the following specific tasks shall be executed:
• Devise a theoretical failure prognosis framework from a system development methodology;
• Synthesize representative and concise mathematical models of the subsea valve actuator and the parameter degradation considering constraining premises;
• Select the required solver algorithms according to the models’ constraints;
• Elaborate test scenarios and evaluate the performance of the im-plemented system with different tunings;
The verification of the proposed solution is partially achieved using experimental data obtained from test-bench experiments – specifically the validation of the UUT model – and partially from exhaustive Monte Carlo simulations using a stochastic model based on laboratory results of an accelerated degradation test.
The result of this work is an effort to directly improve the mainte-nance procedures in the Oil & Gas industry and simultaneously con-tribute with a novel application in the PHM field.
1.6
Methodology
Based in the systematic approach for the implementation of general condition monitoring systems contained in [3], the methodology used to
1.6. METHODOLOGY 11
devise and implement the prognosis system framework in this research is illustrated in a flow chart format in Figure 1.3.
UUT Assessment Failure Assessment Prognosis Approach Testing Performance Evaluation Design & Implemen tation V erific ation & V alidation
Figure 1.3: A systematic approach to implement a generic failure prognosis system (loosely based on [3]).
The methodology is divided into two major parts: Design & Imple-mentation and Verification & Validation, covered thoroughly in chap-ters 3 and 4 respectively. Beginning from the top, the process is re-peated iteratively in sequence until a satisfactory solution is achieved. Vachtsevanos et al. [13] defend a similar designed-in approach, in which the prognosis system is developed iteratively.
The Design & Implementation part is a systematic audition of the case study. The assessment of the UUT comprises the description of its maintenance routine as well as an objective understanding of the pro-cess itself along with its disturbances. The failure assessment involves the study of the degradation processes. Only then with this informa-tion, the methods and algorithms are chosen and tuned according to the particularities of the case. Possibly – as in the case of this case study – some of the prognosis system’s parameters will be narrowed to a range of values, which shall have their influence evaluated through verification & validation.
In the Verification & Validation part, a set of performance metrics have to be selected and a testing procedure is planned. With these
12 CHAPTER 1. INTRODUCTION
definitions the proposed solution can have its performance evaluated. If the solution does not achieve the desired performance, it is verified what assumptions in the Design & Implementation phase have to be refined to improve the results.
The assessment of the UUT and its failures is tackled in the fol-lowing sections. These will provide the conceptual foundation to un-derstand the processes involved and will provide formal inputs in the form of mathematical models to enable the selection of the prognosis approach.
1.7
Structure
This document is divided in five chapters. The first is this introduction, which intended to have provided the reader with the context that moti-vates the development of this study, a general definition of maintenance strategies and failure prognosis systems, the actual state-of-the-art in internal leakage prognosis, as well as what is expected to be achieved from it. It closes with the presentation of the methodology that has been employed to develop the remaining of this dissertation.
The second chapter begins with the theoretical background of sub-sea valve actuators – specifically the design that is the object of this study – followed by a description of their failure distribution and degra-dation mechanisms, considering the operational specifications. In this chapter it is also introduced the mathematical models used for the de-velopment of the prognostic solution.
The third chapter explains the design and implementation process of the presented framework. Initially there is an overview of the prognostic framework devised for the application, highlighting the information flow and the algorithmic perspectives. Then the remaining of this chapter focuses on an in-depth analysis and tuning of the particular methods selected to implement the proposed framework.
In the fourth chapter are shown the verification and validation framework, the simulated scenarios utilized and the different system’s configurations tested. There is an explanation of the performance met-rics in the prognostic field of research that were employed. The results are shown and discussed critically according to each different setting of the system, and a final setting – with best overall results – is elected.
1.7. STRUCTURE 13
Finally the last chapter has a summary wrapping the entire develop-ment and the obtained results. The section ends with a topic regarding the future work necessary for further improvement of this study, with a critical evaluation of the hypotheses and premises made beforehand.
CHAPTER
2
Subsea Valve Actuator and the Prognosis System
Framework
This chapter introduces the process itself – the subsea valve actuator – with an investigation on the distribution of its failure mechanisms. It is then followed by a description of the experiments executed to evalu-ate the degradation mechanism that leads to internal leakage in subsea valve actuators. There is a presentation of the framework for the im-plementation of the prognosis system. Finally, the last part is thorough development of the mathematical models necessary to implement and validate the prognosis system.
2.1
Subsea Valve Actuators
A subsea valve actuator is a drive mechanism to operate and control moving parts in SPS equipment. Typically these actuators are used in all sorts of valves located in subsea Christmas trees, subsea manifolds, and in chokes installed in jumpers [24]. There are different technologies employed in the design of these components. The most traditional designs are direct hydraulic, piloted hydraulic, direct electro-hydraulic, multiplexed electro-hydraulic and, more recently, all-electric systems
16
CHAPTER 2. SUBSEA VALVE ACTUATOR AND THE PROGNOSIS SYSTEM FRAMEWORK
[25].
The valve actuator studied in this research is a novel electro-hydrosta-tic design, shown in Figure 2.1. It was invented at Bosch Rexroth AG and its fundamental design is described in patent WO/2016/023712 [4].
Figure 2.1: The patented concept of the electro-hydrostatic subsea valve actuator [4].
This design is primarily conceived to actuate on gate valves used in subsea Christmas trees in depths of up to 3000 meters, at an average temperature of 4 degrees Celsius. A prototype is under development to operate on gate valves with a stroke length of 80 millimeters. The oper-ational requirements for this prototype define its service life to function within the tolerances for 20 years without maintenance. Throughout this period, it is expected to be able to run one cycle per month, where one cycle is a full extension or retraction of the valve stem.
The design originally contains auxiliary mechanisms for pressure compensation and fail-safe operation. However, a variant in the most simplified form of this concept, is essentially composed by a closed loop hydrostatic system composed of an electrical motor, a positive-displacement pump, a set of pressure piloted check valves and a double-acting hydraulic cylinder. These components are numbered in Figure 2.1 as 51, 50, 75 and 81, and 36 respectively. The focus of this study lies in the piston sealing of the double-acting hydraulic cylinder, located in
2.1. SUBSEA VALVE ACTUATORS 17
the piston of the hydraulic cylinder (number 39 in Figure 2.1). This simplified design shall be considered as the whole actuator throughout the following sections. It is illustrated according to [26] in Figure 2.2, where u designates the electrical input signal of the electrical motor and w is a force disturbance on the cylinder rod.
Motor P ∩ G∩ P∩ u w
Figure 2.2: The simplified electro-hydrostatic valve actuator circuit.
A hydraulic cylinder is a mechanism that converts hydraulic energy into mechanical energy. The pump works the other way around, con-verting rotational kinetic energy into hydraulic energy. The pilot check-valves are check-valves that allow hydraulic flow in only one direction or both directions, depending on the pressure in pilot and inlet ports. The elec-tric motor converts elecelec-tric energy into mechanical energy as well and is the single controlled energy input of this design variant, characterizing this design with an all-electric interface, although ultimately actuating with a hydraulic principle. This design and other variants have been verified and validated in laboratory in the test-bench depicted in Fig-ure 2.3, where the hydraulic cylinder can be seen coated in blue paint in the middle of the picture. The parameters utilized further in this document refer to the components utilized in this test-bench.
18
CHAPTER 2. SUBSEA VALVE ACTUATOR AND THE PROGNOSIS SYSTEM FRAMEWORK
Figure 2.3: The test bench for verification and validation of the novel SVA design.
2.1.1 Failure Distribution of Subsea Valve Actuators
To understand the dynamic processes that develop faults, the mecha-nisms of failure must be assessed. To achieeve this goal, an investiga-tion has analyzed the recorded reliability data of similar components present in this electro-hydrostatic actuator (EHA) design under the same typical environmental and operational settings specified. The data is collected from the Offshore Reliability Data Handbook – the product of a joint industry project constituted by several major Oil & Gas companies in the whole world [27]. This study shows the observed failure distribution according to the classification in [7]. Considering the components in the simplified design shown in Figure 2.2, their
fail-2.1. SUBSEA VALVE ACTUATORS 19
ure distribution is depicted in Figure 2.4 with the failure mechanism designators defined in Table 2.11.
This analysis was conducted during this research and has been pub-lished as a paper and presented in the 10th IFAC SAFEPROCESS Symposium on Fault Detection, Supervision and Safety for Technical Systems 2[5]. 1.0 1.1 1.3 1.6 2.0 2.5 3.1 3.3 5.1 5.2 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
Failure Mechanism Designator
Failure Occurrences [%] Valves, aggregated F ailure Occur rences (× 10 2 %)
Figure 2.4: The failure mechanism distribution for components constituting SVAs [5].
The component defined as "valves, aggregated" in the analysis is understood in this context as being comprised solely by the hydraulic actuating element, which both linear (hydraulic cylinder) or rotary (hy-draulic motor) are aggregated in the OREDA. The research states from the distribution that the most observed failure mechanism is classified
1Other components which don’t figure in the study could not have their failure data assessed, either due to insufficiency of data or due to unsuitable comparison of the operational contexts. These components, however, shall be considered in a future similar analysis utilizing the proprietary Failure Reporting, Analysis and Corrective Action System (FRACAS) database from Bosch Rexroth.
20
CHAPTER 2. SUBSEA VALVE ACTUATOR AND THE PROGNOSIS SYSTEM FRAMEWORK
Table 2.1: Definitions of the failure mechanism designators according to An-nex B of [7].
Failure
Mechanism Description
1.0 Mechanical Failure – General
1.1 Leakage
1.3 Clearance/Alignment Failure 1.6 Sticking
2.0 Material Failure – General
2.5 Breakage 3.1 Control Failure 3.3 Faulty Signal/Alarm 5.1 Blockage 5.2 Contamination as leakage.
The study heuristically deduces – considering the operational and environmental conditions of the system and the literature on the fail-ures of the machine elements involved – that leakage is caused by the degradation of the dynamic seals of the cylinder piston and rod. The mechanical design uses barriers to prevent external leakage caused by the degradation of rod seals, however the leakage caused by the degra-dation of the piston seal – known as internal leakage or cross-port leak-age – still occurs. May et al. [19] stress how the internal leakleak-age alone can greatly decrease the performance of a hydraulic actuator. Further-more, the degradation of the piston seal may ultimately lead to extrude into the gap between the cylinder inner wall and the piston [6]. This abrupt increase in friction may completely hinder the functionality of the cylinder.
2.1.2 Piston Sealing Degradation
The dynamic seals used in hydraulic cylinder pistons are made of elas-tomers, usually taking a toroidal shape, which are designated as O-rings due to its circular cross-section. In this study it is considered only the case of O-rings as sealing element. An O-ring is typically fitted inside a
2.1. SUBSEA VALVE ACTUATORS 21
grove in the piston, according to [28]. A piston may have more than a one seal to mitigate the degradation effects, at the expense of increas-ing the friction load and the risk of extrusion. The sealincreas-ing may also be installed with a pre-loading elastomeric ring underneath to compensate its loss of performance due to degradation.
According to Richter [6], an O-ring has a sealing effect as long as at least one of the following conditions – also depicted in Figure 2.5 – is met:
(i) The O-ring establishes contact throughout the entire clearance due to its ability to deform, caused by the reaction forces of the contact with the cylinder wall and the piston (O-ring inactive). (ii) The O-ring produces a restoring force (pressing against the
clear-ance) which increases proportionally with the pressure load (O-ring active).
Figure 2.5: Sealing effect of an inactive (on the left) and active (on the right) O-ring (adapted from [6]).
The author argues that a leakage is initiated when the O-ring can-not fulfill one of the conditions. This may happen when it is employed outside its operational specifications or due to a mechanism of degra-dation which alters its nominal characteristics. Considering the case of well-dimensioned O-ring – always utilized within the operational spec-ifications – the main degradation processes that may alter its nominal characteristics are compression set, wear, thermal aging, medium aging and manufacturing deviations.
22
CHAPTER 2. SUBSEA VALVE ACTUATOR AND THE PROGNOSIS SYSTEM FRAMEWORK
to its capacity to fulfill its functionality, is its elastic modulus. It is associated with its hardness [29]. According to the generalized Hooke’s law in equation (2.1), the elastic modulus of the elastomer λ is propor-tional to the capacity of the elastomer to withstand deformation due to external stresses.
λ ∝ stress
strain (2.1)
That is, for a given mechanical load, the material will deform pro-portionally with its elastic modulus. As the elastic modulus decreases, the O-ring compresses and no longer provides the restoring force to seal piston clearance. This ultimately leads to an increase of the internal leakage flow. Furthermore, with a sufficiently low elastic modulus, the risk of extrusion becomes greater.
Considering the aforementioned service life specification, operational temperature rating and cycle demands of a subsea valve actuator, the underlying premise considered throughout this study is that the dom-inant failure mechanism acting on the O-ring is the medium aging, while other degradation processes are treated as random disturbances. Medium compatibility is an important issue when selecting the O-ring material and the hydraulic fluid pair. It is well-known the embrittling effect in the elastomer caused by diffused oxygen/ozone in the fluid, however the concentration of these molecules is negligible considering the hydrostatic pressure which the subsea valve actuator is submitted to. A good solution to gather insights on how the medium aging affects the elastomer is to perform an ADT [30]. The standard [31] defines a series of ADTs to evaluate the effects which an elastomeric material suffers from being immersed in a liquid medium, including the change in its elastic modulus.
During the design phase of the subsea valve actuator, a number of samples of the selected O-rings were tested for chemical compatibility with the hydraulic fluids in a series of ADTs according to the aforemen-tioned standard [31]. The theoretical premise is based on the Arrhenius’ law, which establishes the relationship of the effect of the temperature in a reaction rate. It can be written as in equation (2.2). In this equa-tion, the reaction rate R – which is understood here as a ratio of the percentage of reaction per unit of time – is expressed as a function of the temperature T (in Kelvin). R is the universal gas constant, while
2.1. SUBSEA VALVE ACTUATORS 23
the parameters E and G0 depend on the substances involved in the
re-action. Often a deterioration reaction may have multiple intermediary reactions which may even occur in parallel. It is, however, assumed that the reaction-dependent parameters can be adjusted to represent an average of the entire reaction [32].
R(T ) = G0e−RTE (2.2)
The accelerated degradation test described in [31] assumes that the effects of aging (percentage of the reaction) can be reproduced faster by submitting the test subject to a temperature higher than its designed operation temperature Tr (which is assumed to be approximately
con-stant throughout the subject’s service life), while the reaction parame-ters (E and G0) remain constant and independent of the temperature
of the reaction. It defines a conservative fixed test temperature of Ta
= 343K (70 °C). Then dividing the Arrhenius equation evaluated with the design temperature by its counterpart evaluated with the test tem-perature, it’s possible to find a relationship between the test time ta
and the equivalent aging time tr in design conditions, as expressed in
equation (2.3). tr= tae E R( 1 Tr− 1 Ta) (2.3)
Apart from the time itself, the test has been devised to take into account a second factor that influences the aging: the concentration of seawater in the hydraulic fluid. It is considered that seawater may ingress the hydraulic reservoir, slowly mixing with the hydraulic fluid as the system operates during the years. This multi-variable assumption required two test scenarios:
(i) O-Ring accelerated degradation in pure hydraulic fluid.
(ii) O-Ring accelerated degradation in an emulsion of hydraulic fluid with 10% artificial seawater according to [33].
For each case, three test times were selected according to the design service life of the SVA and the recommended intervals in the standard [31]: 168, 504 and 1008 hours. These correspond to an equivalent aging time of 1 year and 8 months, 5 years and 10 years respectively. Each test time has been executed with three new different O-Ring samples.
24
CHAPTER 2. SUBSEA VALVE ACTUATOR AND THE PROGNOSIS SYSTEM FRAMEWORK
Each sample had its mechanical properties measured before and after the tests, which the elastic modulus is the property of interest in this analysis. The design temperature of the SVA is specified as Tr =
277K (4 °C). The activation energy of the reaction has been determined empirically for both test cases. The measured percentage variation of the elastic modulus in respect to its nominal values is shown in Figure 2.6. 100 101 102 103 104 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 1.05 Time (hours)
Percentage of Nominal Elastic Modulus of the O−Ring (%)
Contaminated Fluid (Samples) Pure Fluid (Samples Contaminated Fluid (Fit) Pure Fluid (Fit)
P e rcentage of No minal Elastic Mo dulus of the O-Ring (× 10 2 %)
Figure 2.6: The degradation profile of the elastic modulus of an O-ring im-mersed in different fluids.
The data points can be fairly fitted with a logarithmic regression, from which it is observed that the fluids mixed with seawater heavily accelerate the degradation process in contrast with the pure fluids. Other unmeasurable fluid contaminations that may originate from the produced hydrocarbons or from abrasion of moving parts may also catalyze the degradation.
It has been observed through performance experiments with the de-graded O-rings that at 60% of its nominal elastic modulus, the sealing begins to affect significantly the dynamic performance and ultimately the very functionality of the cylinder, allowing roughly 28% of the inlet flow rate in the cylinder to pass through the leakage gap at
nomi-2.2. PROGNOSIS SYSTEM FRAMEWORK 25
nal operation load. Considering the saturation limits of the actuator, such leakage would bring the system to a state in which the design requirements would no longer be fulfilled within the specified tolerance – namely, the valve closure time – thus it has been determined as the threshold value of this property.
2.2
Prognosis System Framework
As the SVA is a novel design, there is no field data available to allow for a data-driven prognosis method. Nevertheless, the existing experi-mental data of the degradation processes and heuristic knowledge from the operational conditions to enable a model-based approach.
As shown in Figure 2.6, the degradation follows a profile reliably fitted as a logarithmic curve along the time axis, whose slope is affected by a disturbance characterized by the seawater concentration. As hy-pothesized before, the change in internal leakage rate is associated with the elastic modulus of the piston sealing. The model-based prognostic framework intends to track a time-varying parameter (internal leakage resistance) in the observed process – the SVA – fitting it to a model of degradation, and extrapolating this model to the threshold value of the parameter, defined to be θmax= 60%, to estimate a RUL. This
approach requires the following assumptions:
• Other parameters of the process remain time-invariant (constant); • There is a known, univariate and direct relationship between the internal cross-port leakage resistance and the decrement of the elastic modulus of the O-ring due to deterioration;
• There is a known distribution of the random disturbances affect-ing the degradation process and the system;
Considering that the internal leakage can be represented by a time-varying parameter of the observed process model, it can be tracked by employing an adequate parameter estimation algorithm – particu-larly a structured parameter or gray-box estimation technique, since the parameter to be identified is well-defined in the observed process model [34]. Parameter identification techniques, also known as System Identification techniques, require an excitation signal [35]. Therefore,
26
CHAPTER 2. SUBSEA VALVE ACTUATOR AND THE PROGNOSIS SYSTEM FRAMEWORK
it is necessary to interact with the process controller. The valve actua-tors utilized in the Oil & Gas industry already makes use of a monthly excitation procedure called Partial Stroke Test PST to evaluate and maintain the required probability of failure on demand. It consists of a controlled activation of the actuator to move the valve stem back and forth only 5 to 30% of the full stroke. The PST has the advantage of being able to enunciate the dynamics of the actuator relatively well without interfering significantly with the production process [36].
The extrapolation of the deterioration model can be performed us-ing a number of time-series forecastus-ing techniques. In this research, a simple linear Ordinary Least-Squares regression is employed as a means to validate the prognosis framework concept.
In Figure 2.7 the prognostic framework within the process is de-picted in the form of an information flow block-diagram, where the prognosis system itself is contained in the dashed box. The process re-ceives the control signal u from the Controller and is under the distur-bance w. The process output signal y is both used by the Controller for the feedback control loop and by the Process Identification block. The Controller receives reference input r from an external entity – tipically a Master Control System – and the PST reference from the Process Identification block. It gives priority to the external reference input r and schedules the PST to be executed thereafter. The Parameter Esti-mation block periodically sends a PST request rPST to the Controller
and reads both the process control signal u and the process output y (with an additive measurement noise δ) to estimate the internal leak-age parameter ˆθ. The Degradation Model layer reads and stores each estimated parameter ˆθ and fits a model ˆt(ˆθ) of the time in function of the parameter degradation percentage. Finally the Forecast block extrapolates the degradation model to find a RUL estimation ˆtRULand
2.3. MATHEMATICAL MODELS 27 Controller r Process (SVA) w Parameter Estimation ˆ θ Degradation Model ˆ t(ˆθ) Forecast ˆ tRU L± χ rP ST u y + δ
Figure 2.7: An illustration of the information flow for the proposed failure prognostic framework.
There is no means available in this research to validate the proposed prognosis framework in field. This is mainly due to the excessively long periods required for the system to degrade. An ADT of the entire process is not possible, as the system cannot operate under the test temperature. Thus, the proposed solution will be only verified using computer simulation of a model representation of the entire system.
Although the compliance with the OSA-CBM standard is not in the scope of this research, this framework can be viewed within the OSA-CBM structure implicitly in the Data Acquisition and Data Manipula-tion layers, and explicitly in State DetecManipula-tion, Health Assessment and Prognostic Assessment layers. It can be modified later on to comply with the data exchange requirements of the OSA-CBM architecture.
2.3
Mathematical Models
To realize the proposed prognosis framework it is required initially a formal description of the process, its disturbances, and the degradation mechanism itself. In this research, this formal description takes the form of a number of mathematical models. This guides the selection of algorithms and techniques for the parameter estimation and RUL