Study On Soft Fault Diagnosis For Analog Circuits Using Intelligent Optimization

With the rapidly development of electronic technology, in order to shorten the time-to-market of the electronic product and increase its reliability, new theory and fault diagnosis methods need to be established to meet the further requirements of analog circuit fault diagnosis. And, the research on fault diagnosis theories and methods for analog circuits becomes a challenging subject. However, due to the inherent characteristics of analog circuits, such as its nonlinearity and continuous response, continuous variation of the element parameter, tolerances on component parameter, etc, it is difficult for the conventional fault diagnosis theories and methods to achieve the expected diagnosed results in the engineering. Hence, it is important to explore some more efficient diagnosis theories and methods. In this dissertation, on the basis of the fuzzy theory, mathematical programming and particle swarm optimization (PSO), soft fault diagnosis methods for analog circuit are deeply studied. The main works and the contributions are summarized as follows.According to the characteristics of analog circuit faults, based on the direction vector, built by the voltage increment in test nodes, an approach combined fuzzy theory with direction vector to diagnose single soft fault in analog circuit is presented. Firstly, when the tolerance influence is not considered, a linear equation which coefficient matrix is composed by direction vector of voltage increment is built to identify fault in element. From the solution of the equation, the faulty element is located. Then, with the tolerance influence considered, using direction vector of voltage increment in test nodes as fault signature, a fault set is defined. Based on fuzzy math, membership function decided by fixed width approach is used to identify fault state. The given examples show that the proposed method is effective and the diagnosis accuracy is high.Based on the sensitivity analysis, voltage increment equations in test nodes are constructed. The fault diagnosis of analog circuit, through solving the equation, is changed into a decision subject. With the equations as constraint conditions, the problem of diagnosis is transformed into mathematical programming (MP) models and four kinds of MP equations are built. Then, after solving those MP equations, the parameters deviations of each element in diagnosed analog circuit under test (CUT) are evaluated, which enable us to locate the faulty element in CUT. The subjects on faulty element identification and perturbed parameter evaluation are solved together. At the same time, the tolerance question is also handled.With the built voltage increment equations playing the role of a bridge, particle swarm optimization (PSO) is introduced to the fault diagnosis of analog circuit. Firstly, PSO is used to solve the MP equations instead of the commercial software Lingo. After some transformation, PSO is used in fault diagnosis and the dependence on commercial software is solved. Secondly, PSO is connected with wavelet neural network (WNN). By using PSO to optimize the weights in the hidden layers of WNN, PSO-WNN is built and is applied in fault diagnosis. Experimental results show that the method using this PSO-WNN is characterized by many advantages such as faster convergence rate, higher diagnosis rate, etc. and what's more, there is no false diagnosis for existent faults and those new faults can be exactly classified.Through transformation of the voltage increment equations, PSO is applied in the fault diagnosis in analog circuit. The application not only enriches the theory and method of fault diagnosis but expands the PSO's application field. However, as a new kind of intelligence optimization, there are some imperfections in PSO, which need to be improved. In dealing with the question of the relationship between particles behavior and random parameter presented by Kennedy, many researches focus on the effect of the constant value in random parameter and the random number is ignored. In this dissertation, initially, the influence of random parameter on particle's trajectory is studied. The stability analysis of the equation is undertaken, using Lyapunov stability theory. and the result obtained from the analysis offers a more stringent condition on parameters selection in comparison to the existing conclusions. Furthermore, inspired by the effect of individual's experience to group in nature, an improved strategy with a leader particle (PSO-L) is introduced, aiming to find the global optimal solution within a less number of iterations. Through studies with some of the well-known benchmark functions, compared with classical optimizations, the method PSO-L shows rapid convergence ability in the optimization process. Finally, the method PSO-L is applied to diagnose single soft fault in analog circuit. A diagnosis equation is built and PSO methods are used to find the candidate of faulty element. The results show that in the fault diagnosis of analog circuit the improved method remains the merits of basic PSO and owns more advantages. In the end, the effectiveness and practicability of the improved PSO method on the fault diagnosis are tested and verified.