Bayesian Networks in Fault Diagnosis

The following sections are included:

• Preface
• Contents

Fault diagnosis is useful in helping technicians detect, isolate, and identify faults, and troubleshoot. Bayesian network (BN) is probabilistic graphical model that effectively deals with various uncertainty problems. This model is increasingly utilized in fault diagnosis. This chapter presents bibliographical review on the use of BNs in fault diagnosis in the last decades with focus on engineering systems. This work also presents the general procedure of fault diagnosis modeling with BNs; processes include BN structure modeling, BN parameter modeling, BN inference, fault identification, validation, and verification. The chapter provides series of classification schemes for BNs for fault diagnosis, BNs combined with other techniques, and domain of fault diagnosis with BN. This study finally explores current gaps and challenges and several directions for future research.

In order to increase the diagnostic accuracy of ground-source heat pump (GSHP) system, especially for multiple simultaneous faults, this chapter proposes a multi-source information fusion-based fault diagnosis methodology by using Bayesian network (BN) due to the fact that it is considered to be one of the most useful models in the field of probabilistic knowledge representation and reasoning, and can deal with the uncertainty problem of fault diagnosis well. BNs based on sensor data and information observed by human beings are established. Each BN consists of two layers: a fault layer and a fault symptom layer. The BN structure is established according to the cause and effect sequence of faults and symptoms, and the parameters are studied by using noisy-OR and noisy-MAX models. The entire fault diagnosis model is established by combining the two proposed BNs. Six fault diagnosis cases of GSHP system are studied, and the results show that the fault diagnosis model using evidences from only sensor data is accurate for a single fault, but not accurate enough for multiple simultaneous faults. By adding the observed information as evidence, the probability of a fault being present for a single fault of “Refrigerant overcharge” increases to 100% from 99.69%, and the probabilities of fault being present for multiple simultaneous faults of “Non-condensable gas” and “Expansion valve port largen” increase to almost 100% from 61.1% and 52.3%, respectively. In addition, the observed information can correct the wrong fault diagnostic results, such as “Evaporator fouling”. Therefore, the multi-source information fusion-based fault diagnosis model using BN can increase the fault diagnostic accuracy greatly.

Permanent magnet synchronous motor and power electronics-based three-phase inverter are the major components in the modern industrial electric drive system, such as electrical actuators in an all-electric subsea Christmas tree. Inverters are the weakest components in a drive system, and power switches are the most vulnerable components in inverters. Fault detection and diagnosis of inverters are extremely necessary for improving the drive system reliability. Motivated by solving the uncertainty problem in fault diagnosis of inverters, which is caused by various reasons, such as bias and noise of sensors, this chapter proposes a Bayesian network-based data-driven fault diagnosis methodology of three-phase inverters. Two output line-to-line voltages for different fault modes are measured, the signal features are extracted using the fast Fourier transform, the dimensions of samples are reduced using the principal component analysis, and the faults are detected and diagnosed using Bayesian networks. Simulated and experimental data are used to train the fault diagnosis model as well as validate the proposed fault diagnosis methodology.

Bayesian network (BN) is a commonly used tool in probabilistic reasoning of uncertainty in industrial processes, but it requires modeling of large and complex systems in situations, such as fault diagnosis and reliability evaluation. Motivated by the reduction of overall complexities of BNs for fault diagnosis, and the reporting of faults that immediately occur, a real-time fault diagnosis methodology of complex systems with repetitive structures is proposed using object-oriented Bayesian networks (OOBNs). The modeling methodology consists of two main phases: an offline OOBN construction phase and an online fault diagnosis phase. In the offline phase, sensor historical data and expert knowledge are collected and processed to determine the faults and symptoms, and OOBN-based fault diagnosis models are developed subsequently. In the online phase, operator experience and sensor real-time data are placed in the OOBNs to perform the fault diagnosis. According to engineering experience, the judgment rules are defined to obtain the fault diagnosis results.

Transient and intermittent faults of complex electronic systems are difficult to diagnose. As the performance of electronic products degrades over time, the results of fault diagnosis could be different at different times given the identical fault symptoms. A dynamic Bayesian network (DBN)-based fault diagnosis methodology in the presence of transient and intermittent faults for the electronic systems is proposed. DBNs are used to model the dynamic degradation process of electronic products, and Markov chains are used to model the transition relationships of four states, i.e., no fault, transient fault, intermittent fault, and permanent fault. Our fault diagnosis methodology can identify the faulty components and distinguish the fault types. Four fault diagnosis cases of Genius modular redundancy control system are investigated to demonstrate the application of this methodology.

In a complex industrial system, most of the single faults have multiple propagation paths, so any local slight deviation is able to propagate, spread, accumulate, and increase through system fault causal chains. It will finally result in unplanned outages and even catastrophic accidents, which lead to huge economic losses, environmental contamination, or human injuries. In order to ensure system’s intrinsic safety and increase operational performance and reliability in the long term, this study proposes an Integrated Safety Prognosis Model (ISPM) considering the randomness, complexity, and uncertainty of fault propagation.

ISPM is developed based on dynamic Bayesian networks to model the propagation of faults in a complex system, integrating a priori knowledge of the interactions and dependencies among subsystems, components, and the environment of the system, as well as the relationships between fault causes and effects. So, the current safety state and potential risk of system can be assessed by locating potential hazard origins and deducing corresponding possible consequences. Furthermore, ISPM is also developed to predict the future degradation trend in terms of future reliability or performance of the system, and provide proper proactive maintenance plans. Ant colony algorithm is introduced in the ISPM by comprehensively considering two factors — probability and severity of faults — to perform the quantitative risk estimation of the underlining system. The feasibility and benefits of ISPM are investigated with a field case study of the gas turbine compressor system. According to the outputs given by the ISPM in the application, proactive maintenance, safety-related actions, and contingency plans are further discussed and then implemented to keep the system in a high reliability and safety level in the long term.

Integration of a functional hazard operability (HAZOP) approach with dynamic Bayesian network (DBN) reasoning is presented in this chapter. The presented methodology can unveil early deviations in the fault causal chain online. A functional HAZOP study is carried out firstly where a functional plant model (i.e., multilevel flow modeling [MFM]) assisted in the goal-oriented decomposition of the plant’s purpose into the means of achieving that purpose. DBN model is then developed based on the functional HAZOP results to provide a probability-based knowledge representation which is appropriate for the modeling of causal processes with uncertainty. An intelligent fault diagnosis system (IFDS) is proposed based on the whole integrated framework, and investigated in a case study of process plants at a petrochemical corporation. The study shows that the IFDS provides a very efficient paradigm for facilitating HAZOP studies and for enabling the reasoning to reveal potential causes and/or consequences far away from the site of the deviation online.

In complex industrial systems, operating, regulating, maintenance activities, and external incidents occur dynamically and multiple entities in the same or different subsystems interact in a complex manner. Most of the single faults have multiple propagation paths. Any local slight deviation is able to propagate, spread, accumulate, and increase through system fault causal chains. It will finally result in system failure and unplanned outages or even catastrophic accidents. The key issues focus on both how to reduce the probability of fault occurrence and decrease the loss of fault consequence. The implementation of such requirements can be studied in terms of the determination of the fault root causes, prediction of the possible consequence, and also estimation of the risk and timing of various maintenance activities, which are considered in a failure prognosis scheme.

This study proposes a DBN-based failure prognosis method for complex systems. Not only the interaction between components, but also the influence of the layers of protection in the system is considered when the dynamic failure scenarios are analyzed. Therefore, the proposed method considers multiple factors including degradation mechanism, parameter deviation, the response of the layers of protection, and also the external environment. With this model, the dynamic influence diagram of the components' degradation trends can be calculated and used to evaluate the different effects of the layers of protection quantitatively. Some key problems are also explained such as how to determine the new nodes in DBN representing the behavior of protective layers and how to update the conditional probability table in the extended model.

In the case study, the proposed method is tested on the flue gas energy recovery system (FGERS) which is widely used in the petrochemical industry to demonstrate its effectiveness. It is of great help for early warning and optimization of the layers of protection in complex industrial systems.

Fault diagnosis for a solar-assisted heat pump (SAHP) system in the presence of incomplete data and expert knowledge is discussed in this chapter. A method for parameter learning of Bayesian networks (BNs) from incomplete data based on the back-propagation (BP) neural network and maximum likelihood estimation (MLE), which is called BPMLE method, is presented. The BP neural network is utilized to impute the missing data and the complete data sets are addressed with MLE to obtain the parameters of BN. A method for parameter estimation under incomplete expert knowledge based on BP neural networks and fuzzy set theory is also presented, which is called BP-FS method. Similarly, the missing information is imputed by the trained BP neural network. Fuzzy set theory is employed to quantify the parameters of BN based on the complete qualitative expert knowledge. The presented methods are applied to parameter learning of diagnostic BN for an SAHP system with incomplete simulation data and expert knowledge. The developed BN can perform fault diagnosis with complete or incomplete symptoms.

In this chapter, a novel approach of developing the Bayesian network for fault diagnosis based on operation procedures is presented. The proposed Bayesian network consists of operation procedure layer, fault layer, and fault symptom layer. First, operation procedure layer containing procedure nodes and state decision nodes is developed. Second, the fault layer is determined based on the state decision nodes in the operation procedure layer. Then fault symptoms sensitive to the concerned faults are developed. Finally, the entire Bayesian network is established by integrating the three layers. The presented approach is applied to the hydraulic control system of subsea blowout preventer (BOP). Taking an example of closing the BOP, the operation procedures are illustrated. The entire Bayesian network for the fault diagnosis of closing the BOP is established. Several cases possible to appear during the closing process are studied to evaluate the developed model.

Drilling operations of offshore oil and gas fields are characterized by high technical complexities, high risks, and high costs, since they are always in harsh environments with complicated geological factors. Lost circulation or well “kick” is a typical hazardous event that may occur while drilling wells and it also may develop into a blowout accident without being well handled. It is necessary to identify and analyze the root causes of these events and their consequences in order to prevent serious accidents from happening. In a drilling operation, the risk of blowout may change with time, depending on the operation stage, and such kind of dynamics should be captured in risk assessment. This chapter presents an approach for determining the conditional probabilities of hazardous events and their consequences. The approach includes models that take into account the influence of degradation and (if applicable) new real-time information which represents the changing model parameters (such as state change of mud density). Such an approach is based on the dynamic Bayesian network (DBN) theory and then incorporates additional nodes to address the model uncertainties and parameter uncertainties. In addition, the effect of degradation, which had been ignored in the existing researches, is also taken into account. Given that a hazardous event has occurred, this presented model can be used to predict the risk evolution, as well to reason its root causes during the offshore drilling operation. A bowtie model is established to link the potential incident scenarios with the pressure regimes and formation load capacity, and then the model is translated into a DBN. DBN inference is adapted to perform prediction and diagnosis for dynamic risk assessment, and then a sensitivity analysis is carried out to find the relative importance of each root cause. A case study focusing on lost circulation during three drilling scenarios is adapted to illustrate the feasibility of the proposed approach.

This chapter proposes a fault diagnosis methodology for a gear pump based on the ensemble empirical mode decomposition (EEMD) method and the Bayesian network. Essentially, the presented scheme is a multisource information fusion-based methodology. Compared with the conventional fault diagnosis with only EEMD, the proposed method is able to take advantage of all useful information besides sensor signals. The presented diagnostic Bayesian network consists of a fault layer, a fault feature layer, and a multisource information layer. Vibration signals from sensor measurement are decomposed by the EEMD method and the energy of intrinsic mode functions (IMFs) are calculated as fault features. These features are added into the fault feature layer in the Bayesian network. The other sources of useful information are added to the information layer. The generalized three-layer Bayesian network can be developed by fully incorporating faults and fault symptoms as well as other useful information such as naked eyes’ inspection and maintenance records. Therefore, the diagnostic accuracy and capacity can be improved. The proposed methodology is applied to the fault diagnosis of a gear pump and the structure and parameters of the Bayesian network are established. Compared with artificial neural network and support vector machine classification algorithms, the proposed model has the best diagnostic performance when only sensor data is used. A case study has demonstrated that some information from human observation or system repair records is very helpful for fault diagnosis. It is effective and efficient in diagnosing faults based on uncertain, incomplete information.

The following section is included:

• Index