Venkat Venkatasubramanian

A review of process fault detection and diagnosis: Part I: Quantitative model-based methods

Fault detection and diagnosis is an important problem in process engineering. It is the central component of abnormal event management (AEM) which has attracted a lot of attention recently. AEM deals with the timely detection, diagnosis and correction of abnormal conditions of faults in a process. Early detection and diagnosis of process faults while the plant is still operating in a controllable region can help avoid abnormal event progression and reduce productivity loss. Since the petrochemical industries lose an estimated 20 billion dollars every year, they have rated AEM as their number one problem that needs to be solved. Hence, there is considerable interest in this field now from industrial practitioners as well as academic researchers, as opposed to a decade or so ago. There is an abundance of literature on process fault diagnosis ranging from analytical methods to artificial intelligence and statistical approaches. From a modelling perspective, there are methods that require accurate process models, semi-quantitative models, or qualitative models. At the other end of the spectrum, there are methods that do not assume any form of model information and rely only on historic process data. In addition, given the process knowledge, there are different search techniques that can be applied to perform diagnosis. Such a collection of bewildering array of methodologies and alternatives often poses a difficult challenge to any aspirant who is not a specialist in these techniques. Some of these ideas seem so far apart from one another that a non-expert researcher or practitioner is often left wondering about the suitability of a method for his or her diagnostic situation. While there have been some excellent reviews in this field in the past, they often focused on a particular branch, such as analytical models, of this broad discipline. The basic aim of this three part series of papers is to provide a systematic and comparative study of various diagnostic methods from different perspectives. We broadly classify fault diagnosis methods into three general categories and review them in three parts. They are quantitative model-based methods, qualitative model-based methods, and process history based methods. In the first part of the series, the problem of fault diagnosis is introduced and approaches based on quantitative models are reviewed. In the remaining two parts, methods based on qualitative models and process history data are reviewed. Furthermore, these disparate methods will be compared and evaluated based on a common set of criteria introduced in the first part of the series. We conclude the series with a discussion on the relationship of fault diagnosis to other process operations and on emerging trends such as hybrid blackboard-based frameworks for fault diagnosis.

A review of process fault detection and diagnosis: Part III: Process history based methods

In this final part, we discuss fault diagnosis methods that are based on historic process knowledge. We also compare and evaluate the various methodologies reviewed in this series in terms of the set of desirable characteristics we proposed in Part I. This comparative study reveals the relative strengths and weaknesses of the different approaches. One realizes that no single method has all the desirable features one would like a diagnostic system to possess. It is our view that some of these methods can complement one another resulting in better diagnostic systems. Integrating these complementary features is one way to develop hybrid systems that could overcome the limitations of individual solution strategies. The important role of fault diagnosis in the broader context of process operations is also outlined. We also discuss the technical challenges in research and development that need to be addressed for the successful design and implementation of practical intelligent supervisory control systems for the process industries.

A review of process fault detection and diagnosis: Part II: Qualitative models and search strategies

In this part of the paper, we review qualitative model representations and search strategies used in fault diagnostic systems. Qualitative models are usually developed based on some fundamental understanding of the physics and chemistry of the process. Various forms of qualitative models such as causal models and abstraction hierarchies are discussed. The relative advantages and disadvantages of these representations are highlighted. In terms of search strategies, we broadly classify them as topographic and symptomatic search techniques. Topographic searches perform malfunction analysis using a template of normal operation, whereas, symptomatic searches look for symptoms to direct the search to the fault location. Various forms of topographic and symptomatic search strategies are discussed.

Process fault detection and diagnosis using neural networks—I. steady-state processes

Abstract An analysis of the learning, recall and generalization characteristics of neural networks for detecting and diagnosing process failures in steady state processes is presented. The single fault assumption has been relaxed to include multiple causal origins of the symptoms. The effect of incomplete and uncertain process symptom data such as sensor faults, and the effect of degradation of different hidden units, on the performance of the network, have been analyzed. Various neural network topologies (i.e. number of hidden units and hidden layers) have been tested and compared. The results show that accurate recall and generalization behavior is observed during the diagnosis of single faults. Performance during recall improves at first with an increase in the number of hidden units and with the amount of training, and then attains convergence. In general, performance during generalization improves with the extent of training. The networks are also able to diagnose correctly even in the presence of faulty operation of certain sensors. Networks trained on single faults are able to accurately diagnose measurement patterns resulting from multiple faults in a large majority of the cases studied. Graceful degradation of diagnostic function was observed in many of the multiple-fault cases that were not accurately diagnosed.

Computer-aided molecular design using genetic algorithms

Abstract Designing new molecules possessing desired properties is an important activity in the chemical and pharmaceutical industries. Much of this design involves an elaborate and expensive trial-and-error process that is difficult to automate. The present study describes a new computer-aided molecular design approach using genetic algorithms. Unlike traditional search and optimization techniques, genetic algorithms perform a guided stochastic search where improved solutions are achieved by sampling areas of the parameter space that have a higher probability for good solutions. Moreover, genetic algorithms allow for the direct incorporation of higher level chemical knowledge and reasoning strategies to make the search more efficient. The utility of genetic algorithms for molecular design is demonstrated with some case studies in polymer design. The merits and potential deficiencies of this approach are also discussed.

A syntactic pattern-recognition approach for process monitoring and fault diagnosis

Abstract Process operators often deal with vast amounts of sensor data that are typically updated every few minutes. From such real-time data, operators extract interesting and important qualitative trends and features that describes the essential aspects of the process behavior. This level of understanding is essential for performing causal reasoning about process behavior. To aid this decision-making process of operators, a syntactic pattern-recognition approach for process monitoring has been developed. The syntactic pattern-recognition approach has two main parts: (i) a set of primitives that form the trend description language to represent basic changes in trends and (ii) a grammar to perform error correction and explanation generation. The syntactic approach to process monitoring provides a capability to describe complex patterns using a small set of simple primitive patterns. In this work, a backpropagation-based neural network was trained to identify the presence of the appropriate primitives in a trend of noisy process data. A process grammar which can utilize both contextual and non-contextual information to perform error correction and explanation generation has also been developed. These are discussed with the aid of a FCCU case study.

Intelligent systems for HAZOP analysis of complex process plants

Abstract Process safety, occupational health and environmental issues are ever increasing in importance in response to heightening public concerns and the resultant tightening of regulations. The process industries are addressing these concerns with a systematic and thorough process hazards analysis (PHA) of their new, as well as existing facilities. Given the enormous amounts of time, effort and money involved in performing the PHA reviews, there exists considerable incentive for automating the process hazards analysis of chemical process plants. In this paper, we review the progress in this area over the past few years. We also discuss the progress that has been made in our laboratory on the industrial application of intelligent systems for operating procedure synthesis and HAZOP analysis. Recent advances in this area have promising implications for process hazards analysis, inherently safer design, operator training and real-time fault diagnosis.

Model-based reasoning in diagnostic expert systems for chemical process plants

Abstract A prototype expert system, called MODEX, for locating the cause(s) of a set of abnormalities in a chemical process id described. We discuss a methodology that aids the developement of expert systems which are process-independent, transparent in their reasoning, and capable of diagnosing a wide diversity of faults. The domain knowlege of the system is based on qualitative reasoning principles and captures physical interconnections between equipment units as well as causal relationships among process state variables. The inference strategy uses model-based reasoning for analyzing the plant behavior. Using a variant of the technique adopted from fault tree synthesis, an initially observed abnormal symptom is considered to be a top level event and a tree structure is constructed as the system searches for a basic event to which the fault can be traced. The diagnostic reasoning process is driven by a problem reduction strategy. The knowledge base is process-independent, thereby enhancing the generality of the expert system. Reasoning from first-principles with the aid of causal and fault models facilitates the diagnoses of novel or unanticipated faults. The system does not assume a single causal origin for all initially observed faults in the chemical process. Moreover, the system has the ability to locate multiple basic causes of a fault. The methodology also permits one to investigate the causal origins of multiple, unrelated, faults. The system provides explanations to user queries at various degrees of detail. Two test cases are discussed in detail.

Challenges in the industrial applications of fault diagnostic systems

Process fault diagnosis (PFD) involves interpreting the current status of the plant given sensor readings and process knowledge. Early diagnosis of process faults while the plant is still operating in a controllable region can help avoid event progression and reduce the amount of productivity loss during an abnormal event. PFD forms the first step in abnormal situation management (ASM), which aims at timely detection, diagnosis and correction of abnormal conditions. However the problem of PFD is made considerably difficult by the scale and complexity of modern plants. We briefly outline the various challenges in the area of PFD and review the existing methods to tackle them. We argue that a hybrid blackboard-based framework utilizing collective problem solving is the most promising approach. The efforts of the ASM consortium in pursuing the implementation of the state-of-the-art technologies at plant sites are also described.

PCA-SDG based process monitoring and fault diagnosis

Abstract Significant research has been done in recent years to use principal component analysis (PCA) for process fault diagnosis. The general approach involves manual interpretation of measured variable contributions to the residual and/or principal components. For a large chemical process, this could be tedious and often impossible. In addition, it hampers the automation of high-level analysis and decision support tasks that require root cause information. In this work, the interpretation of PCA-based contributions is automated using signed digraphs (SDGs). Also, a serious limitation of SDG-based diagnosis – the assumption of a single fault – is overcome by developing a SDG-based multiple fault diagnosis algorithm. The implementation of the PCA-SDG-based fault diagnosis algorithms is done using G2. Its application is illustrated on the Amoco Model IV Fluidized Catalytic Cracking Unit (FCCU).