The fault diagnosis technology for power electronic circuits includes two aspects: 1. Fault information detection: using certain detection techniques to obtain the required fault information when a fault occurs, for fault analysis and reasoning; 2. Fault diagnosis: based on the detected fault information, using appropriate fault diagnosis methods to analyze and reason about the fault, identify the cause of the fault, and locate the fault location. Traditional fault diagnosis methods are also widely applied in power electronic circuit fault diagnosis, such as fault dictionary method, fault tree, expert systems, etc. (1) Fault Dictionary Method. Store a set of typical measured characteristic values and fault values in a table format, and judge faults by comparing measured values with characteristic values. First, simulate the normal state and all hard fault states of the circuit using a computer to establish a fault dictionary. Then analyze the port tests to identify faults, that is, compare the voltages measured at selected nodes with those in the fault dictionary, and use certain isolation algorithms to find the corresponding fault.
The fault dictionary method has great practical value for fault diagnosis of analog and digital circuits, but it can only solve single fault diagnosis; the combination of multiple faults is large and difficult to implement in practice.
(2) Fault Tree Method. The fault tree diagnosis method analyzes various factors that may cause system failure (including hardware, software, environment, human factors), draws a logic block diagram, i.e., the fault tree, and searches from the top event of the fault tree to find the cause of the fault. The fault tree expresses the internal relationships of the system and indicates the logical relationship between component faults and the system.Current Transformer On-site Tester Solution solves the problems of high work intensity and cumbersome operation in on-site verification of current transformers and voltage transformers, and the product is reliable and powerful.
The fault tree diagnosis is intuitive, flexible, and universal, but building the tree is labor-intensive, cumbersome, and error-prone, and it is suitable for problems with a small fault space.
(3) Residual Method. The residual method is a fault diagnosis method based on analytical models. It performs online fault diagnosis and fault location of the main circuit of power electronic devices by studying the residual between the actual system and the reference model's characteristic output. This method is also applicable to fault diagnosis of inverter main circuits. The reference model method for fault diagnosis of power electronic circuits has the characteristics of fewer measurements, simple criteria, and independence from output magnitude. Especially in fault diagnosis of complex power electronic circuits, this method has more obvious advantages.
(4) Method of Directly Detecting Voltage Across Power Devices or Bridge Arm Current. By detecting the voltage across each power device or the current of each bridge arm, the working mode of the power device is obtained, and then compared with the trigger pulse in sequential logic to determine whether the diagnosed object is faulty. This method requires detecting the voltage and current of each diagnosed device, requiring many measurement points and dedicated detection and logic circuits. This method can also achieve fault diagnosis by measuring the input and output of the circuit. During normal operation, the input and output of the circuit vary within a certain range; when it exceeds this range, it can be considered that a fault has occurred. Additionally, it is possible to judge whether a fault has occurred by measuring whether the rate of change of input and output variables exceeds the range. Although this method is simple, it has poor anti-interference capability.
(5) Expert System Diagnosis Method. An expert system is an intelligent computer program system designed by utilizing computer reasoning capabilities, rich experience of domain experts, internal causal relationships of the system, and machine learning functions of artificial intelligence to solve complex system fault diagnosis problems. Expert systems formally describe empirical diagnostic knowledge, break through personal limitations for widespread dissemination, facilitate the storage and promotion of expert experience, give full play to the role of specialized talents, and open a new way to comprehensively utilize expert knowledge. They are more reliable, flexible, and unaffected by the environment than human experts. In the knowledge structure of an expert system, the knowledge base and inference control are relatively independent, can be rewritten, added, or deleted, and can be combined with other diagnostic methods to form applications with a knowledge structure, possessing human-machine joint diagnosis functions, fully leveraging the subjective initiative of on-site technical personnel, and gradually accumulating experience to become more perfect. Therefore, it is a very viable fault diagnosis method.
The basic idea of expert system diagnosis is: first, establish a reliable knowledge base through experiments or simulations, which includes circuit environment knowledge, system knowledge, and a rule base. The knowledge base reflects the causal relationships of the system, specifically in the fault diagnosis system, the causal relationships between system variables and fault types and fault points; then obtain the characteristic variable values in actual operation through the human-machine interface; apply them to the rule base for reasoning, and obtain the basic working state and fault information of the circuit. The disadvantage of this method is that establishing the knowledge base is difficult, especially when the knowledge base is large. Traditional fault diagnosis expert systems are mostly rule-based expert systems, which compile domain knowledge into a series of production rules (in the form of IF..., THEN...). Such expert systems can solve fault diagnosis problems for many systems, but because complex systems require a large number of production rules (which mainly rely on manual writing), fault diagnosis expert systems run slowly and are difficult to adapt to real-time environment requirements; in addition, when encountering new faults or new information not seen before, such established expert fault diagnosis systems often cannot handle them correctly, and may suffer from "matching conflicts", "combinatorial explosion", and "infinite recursion" due to weak reasoning ability. To solve these problems, unless rules are continuously updated, but new rules may be intertwined with existing rules, which inevitably leads to difficulties in adding and deleting rules. In summary, expert diagnosis systems have disadvantages such as the "bottleneck" of knowledge acquisition, difficulty in maintenance, narrow application scope, weak diagnostic ability, and inability to adapt to fuzzy problems.
