intelligent subsea control

Transcription

1 40 SUBSEA CONTROL How artificial intelligence can be used to minimise well shutdown through integrated fault detection and analysis. By E Altamiranda and E Colina. While there might be topside, there are currently no tools for early fault detection and diagnosis in the subsea control arena intelligent subsea control BP THE NEED TO operate industrial plants in a smooth and sound manner to ensure compliance with technical specifications and safeguard product quality is heightened in today s increasingly competitive globalised economy. Competitive advantage can be gained by reducing raw material and energy consumption and energy consumption costs, maximizing plant throughput and by meeting rigorous environmental regulations. The need for superior economic control of plant operation is presently supported by the general availability of modern hardware and software resources in industrial processes. More specifically the widespread adoption of Distributed Control Systems (DCSs) allows for data acquisition and control strategy implementation according to plant design specifications. One of the important challenges facing control system engineers is how to design and implement intelligent systems that may assist supervision and decision making such as abnormal situation management (ASM), start up and shut down, controller performance assessment and so on. In engineering, supervision becomes more and more important in order to improve reliability and safety. The classical method is based on monitoring important measurable variables followed by alarm triggering if certain tolerances are exceeded. This limit value based monitoring is simple and reliable but it does not solve the task of early fault detection and diagnosis. The experience of the last 20 years has shown that earlier fault detection can be reached by gathering information, especially by using the relationships between several measurable quantities in terms of mathematical process s (by using analytical redundancy). Heuristic information such as human operator observations or process expert experience is also very relevant for diagnosis. A complete fault diagnosis therefore requires the systematic treatment of both analytical and heuristic.

2 Over the last decade, there has been a huge increase in the application of subsea production systems for the production of oil and gas from subsea wells 41 ARTIFICIAL INTELLIGENCE The use of artificial intelligence approaches such as neural networks, genetic algorithms and fuzzy logic has been increasing in the last few years. These approaches have potential for developing reasoning strategies for applications such as fault detection, diagnosis, supervision and decisionmaking among others. These techniques have been successfully applied in other oil industry related businesses such as refinery and petrochemical where the processes and control systems involved have a very high complexity. Presently there are no tools for early fault detection and diagnosis in the subsea control systems area. Troubleshooting and fault finding are normally performed with procedures when the fault is already affecting the system performance and, in most cases finding the root cause of the fault is not straight forward. This fact affects subsea operations and does not provide any support for maintenance plans. Intelligent supervision, fault detection and diagnosis techniques successfully applied to other oil industry related processes are perfectly applicable to subsea technology. SUBSEA PRODUCTION CONTROL Over the last decade, there has been a huge increase in the application of subsea production systems for the production of oil and gas from subsea wells. The subsea system comprises a wellhead, valve tree (xmas tree) equipment, pipelines, structures and piping systems among others. In many cases a number of wellheads have to be controlled from a single location. The control system provides operation of valves and chokes on subsea completions, templates, manifolds and pipelines. The design of a control system must also provide a means for safe shutdown on failure of equipments or loss of electrical/hydraulic control from the topside (a platform or floating facility) and other safety features that automatically prevent dangerous events. FAULT DETECTION AND DIAGNOSIS An overall scheme of a knowledge based integrated fault detection and diagnosis system is presented in the accompanying diagram. The main tasks can be subdivided in fault detection by analytic and heuristic generation and fault diagnosis ANALYTIC SYMPTOMS GENERATION The analytical knowledge on the process is used to produce quantifiable, analytical information based on measured process variables and data processing. This information is used to generate first CONTROL EQUIPMENT Topside Topside control system equipment comprises a hydraulic power unit (HPU), an electrical power unit (EPU) and a well control panel or master control station (MCS). The HPU provides high and lowpressure hydraulic supplies and is usually powered by electric motors, although redundancy is sometimes provided by air drives. The HPU includes tanks, pumps, a contamination control system and hydraulic control valves. A programmable logic controller PLC or PC based EPU may be integrated with the platform control system or it may be a selfcontained unit. Umbilicals An umbilical is a conduit between the topside host facility and the subsea control system and is used for chemical and/or hydraulic fluids, electric power and electric control signals. The hydraulic power control lines are individual hoses or tubes manufactured from steel or thermoplastic materials and encased by the umbilical outer sheath. The electrical control cables supplying power and control can be either bundled with hydraulic lines or laid separately. Subsea The production control system provides control of all functions of the subsea production system. Conventionally, subsea functions include operation and control of: down hole, safety valves, subsea chokes, production valves mounted on the xmas tree and utility functions such as monitoring of fluid characteristics, pressure leakage, valve positions, etc. Control System Configurations Distances between top side production facilities and subsea installations have generally increased. Due to both multiple well developments and water depth, early methods using direct hydraulic control of subsea valves have become less feasible because of the operational limitations and the size and cost of multicore umbilicals required to provide hydraulic and power transmission. This has led to the development of more advanced and complex control methods using piloted hydraulic systems, sequential piloted systems and electrohydraulic systems (hardwired and multiplexed). The complexity and performance characteristics of subsea control systems depend on the control configuration used. The selection is associated predominantly with technical factors like distance between control points (offset distance between the platform and the trees), water depth, required response speed during execution of subsea functions and type of subsea installations (single or multiple wellheads). The significant costs associated with the design, manufacture and installation of subsea control systems have provided incentives to improve the existing systems and to invest in research for new systems. Some other trends still in development are the subsea powered autonomous remote control systems (SPARCS) and the integrated control buoy. These reduce the overall cost of control systems by removing the need for umbilical and topside equipment required for conventional systems. August/September 2007 Control & Automation

3 42 SUBSEA CONTROL characteristic values by 1) Limit value checking of direct measurable signals. Characteristic values are exceeded signal tolerances. 2) Signal analysis of direct measurable signals, using signal s like correlation functions, frequency spectra, autoregressive moving average (ARMA). Characteristic values are, for example, variances, amplitudes, frequency or parameters. 3) analysis by using mathematical process s together with parameter estimation, state estimation and parity equation methods. Characteristic values are parameters, states variables or residuals. In some cases physical features can be extracted from characteristic values, such as physical defined process coefficients or special filtered or transformed residuals. These features are then compared with the normal features of the nonfaulty process. For this, methods of change detection and classification are applied. As analytic the resulting changes (discrepancies) of the described direct measured signals, signal s or process s are considered. HEURISTIC SYMPTOMS GENERATION In addition to the generation with quantifiable information, heuristic can be produced from human experts and operators, and observations and inspections The process history in the form of performed maintenance, repair, former faults, life time, load measures constitute a source of heuristic information. Statistical data achieved from experience with the same or similar processes can be added. FAULT DIAGNOSIS The task of fault diagnosis consists of determining the type, size and location of the fault, as well as its time of detection based on the observed analytical and heuristic. With the aid of heuristic kno wledge in the form of heuristic process s (qualitative s), fault, causalities and weighting of effects different diagnostic reasoning strategies can be applied. KNOWLEDGE REPRESENTATION In most applications, a certain amount of knowledge about the behaviour is present. This should be exploited when a diagnostic system is created. Even if exact values for thresholds are not known, there usually is some insight about the process, such as physical understanding of similar faults or similar effects of faults on certain. Furthermore, the selection of the for the diagnosis becomes a matter of robustness. The fault diagnosis, therefore, must be based on the appropriate subset of all available. Different faults can be categorized into larger groups if their effect on the process is similar. It is then advantageous to find a classification system for the larger groups first and later separate the faults within them. This leads to the concept of a hierarchical diagnosis system Artificial intelligent techniques such as neural networks and fuzzy logic have been successfully used for fault classification. Neural networks capabilities such as learning, adaptation and information distributed processing hybridized with the linguistic representation capabilities for the description of a cause symptom relationship of fuzzy logic methods have provided excellent potential to represent heuristic and analytical behaviour in an integrated way for this kind of applications. Fault tree methods have also been hybridized with intelligent algorithms for hierarchical diagnosis systems. INTELLIGENT SUPERVISION Some difficulties associated Analytical knowledge Analytical process Filtering estimation Change detection classification Normal features Heuristic process Fault causalitics Weighting of effects Heuristic knowledge Analytic generation Data processing Feature extraction Measured variables Characteristic values Features Change detection classification Statistical evaluation Forward chaining Faults Analytic Observed variables Backward chaining Fig 1: Integrated Fault Detection and Diagnosis Scheme Supervisory control level (Discrete control) Discrete control patterns Translator Regulatory control level (Continuous control) Fig 2: Supervisory Control Scheme Input variables Automaton Unified Diagnosed faults Continuous control patterns (Set points) Regulatory control Subsea process and control system Hydraulic Communication Output variables Electrical Electronic Fig 3: Residuals Generation for Fault Detection Scheme Sequencing control patterns Operational decisions and warning messages for mantainance tasks Translator Discrete control patterns Fig 4: Intelligent Supervisory System Managment level Supervisor (automaton) Operator observations history statistics Heuristic United representation Deviations Fault decision Heuristic generation detector Integration of information Fault diagnosis (inference reasoning) variables information Instrumentation Integrated fault detection and diagnosis system Subsea process and control system r 1 r 2 r 3 r 4 r 5 Residuals Supervisory level (Discrete event domain) Intelligent event detector Continuous process level

4 Artificial intelligent techniques such as neural networks and fuzzy logic have been successfully used for fault classification 43 with the supervision of complex dynamic processes are usually related to multiple operational domains due to the presence of nonlinear phenomena and unpredictable or partially known disturbances, which affect the process performance. In order to alleviate such difficulties, it is important to incorporate strategies for different operational domains. Including adaptive capacities to deal with uncertain situations in order to allow the coordination of distributed controllers and decision making support for a satisfactory task assignment. A large number of industrial processes operate on a continuous time base, and they are usually described in terms of differential equations. Other types of processes, of discrete time nature, may be represented using transition systems, for example the sequential operation in the automotive manufacturing, chemical processes where batch operation is involved and changes in operational regions related to a continuous process are usually described as discrete systems. In transition systems based s; the process under scrutiny is described in terms of discrete events. The dynamic systems whose behaviour depends on the interaction between continuous time processes and discrete controllers are called hybrid systems. In the process control area, the continuous time process of a hybrid system corresponds to the physical process itself, which must be controlled. A discrete event system, on the other hand, represents a supervisor (automaton), which reacts in the presence of generated events from the continuous time process in order to fulfil system specifications or support decision making tasks. It is considered the supervisory control scheme philosophy presented in which facilitates the decision making task related to control actions for improving the operation of complex dynamics processes that may be composed of interconnected s, characterised by different operational conditions and usually subjected to external disturbances. The scheme is structured in two layers, named regulatory control level and supervisory control level. The regulatory control level is related to the process dynamics and is in charge of generating direct control action to be applied to the process. This level is governed by a supervisor (automaton) in the supervisory level, which assigns discrete control patterns, based on generated process events. The events related to the continuous level may indicate changes in the process behaviour or changes in the current operational region that could impede the achievement of required process specifications. The supervisory level contains a process representation based on operational regions and transitions among them. The event detector characterises the process behaviour and evaluates the performance August/September 2007 Control & Automation

5 44 SUBSEA CONTROL The s are highly interconnected and a fault generated in one of them can be propagated to other s affecting the whole control system performance, subsea operation and production of the regulatory control system monitoring the deviations between process variables and assigned set points for basic controllers. This information is translated to the automaton in terms of discrete events in order to yield discrete control patterns, which will be codified by a translator to the regulatory control system in terms of appropriate set points for basic controllers, alarms and messages to operators. SUPERVISION, DETECTION AND DIAGNOSIS Conventional subsea production control systems, with multiplexed electrohydraulic configuration can be subdivided into the main following s: Electrical s. Comprising the power supply and electrical distribution system; the hydraulic including the hydraulic supply, hydraulic distribution and the subsea hydraulic sub process inside the subsea control modules; the electronic comprising the electronic within the subsea control modules and surface control equipment, the communication which comprises all the variables that determine the communication quality which are mainly connected with the electrical when communication on power is used and the instrumentation system including the external and internal sensors and actuators on the xmas trees and the control modules respectively. These s are highly interconnected and a fault generated in one of them can be propagated to other s affecting the whole control system performance, subsea operation and production. Hence the importance of developing integrated fault detection, diagnosis methods and appropriate supervision tools to be able to perform in a more efficient manner, troubleshooting and fault identification to support subsea operation decision making and maintenance tasks. Diagnostic s. Corresponding to the nominal conditions for each have to be achieved to generate the analytical. Cross interactions among s must be included to obtain a reliable multivariable. Fig 3 represents a residuals generation scheme for fault detection incorporating the different s and the main interaction between them. FAULT DIAGNOSIS The heuristic generation considers process observation from experts and operators, process history and process data. Then analytical and heuristic are integrated according to the scheme presented in Fig 1. A fault diagnosis task will consider the heuristic knowledge and the unified according to the diagram presented in Fig 1. Intelligent techniques have to be incorporated for classification and fault identification as was mentioned earlier. INTELLIGENT SUPERVISION The proposed supervision scheme is based on the philosophy presented however, for this application, that the discrete control patterns will provide information to support appropriate decision making for subsea operations and warning messages to support troubleshooting and maintenance tasks. Fig 4 illustrates the integrated supervision scheme. The continuous process level represents all the production control system which interacts with the fault detection and diagnostic system to generate the faults identification when they are produced. This information is processed by an intelligent event detector, which allows mapping the identified faults in discrete events. The discrete events are used in the supervisory system in order to generate the appropriate discrete control patterns for supporting troubleshooting, decision making and maintenance tasks. These patterns have to be sequenced since it is a multivariable process with several s. The translator block will translate the discrete control patterns in specific decisions and messages for subsea operation and maintenance tasks. The supervisory system must interact with management levels to be able to update the operational regions in accordance with management priorities. CONCLUSION The above approach provides decision making support for subsea operations and maintenance tasks (preventive maintenance) and also provides more efficient mechanisms for troubleshooting when faults and events are generated. It suggests the incorporation of each (hydraulic, electrical, electronic, communication and instrumentation) with the corresponding interactions for the diagnostic s in the fault detection and diagnostic scheme. Information from management levels is also highly valuable to ensure the supervisor can generate the appropriate discrete patterns according to operational and management priorities.