Research On Fault Diagnosis Method Of Analog Circuits Based On Information Theory

With the rapid development of modern electronic technology, the demand of research on system performance, such as reliability, linearity, accuracy, security and self-healing, has been urgently raised. Especially, the demand of research on diagnosis of circuits with high performance is also becoming more and more appealing. The analog component has many great characters such as good stability, multiple choices, low power-consuming and so on, so it is widely used in many fields such as automatic control, guidance, electronic countermeasure and microwave communication. It is noted that the analog subsystems, which are composed by analog components, plays a fundamental role in the whole system. The stability and testability of such analog subsystem have a critical influence on the optimal performance of the whole system. As a result, the diagnosis of analog circuits is the key technology for both construction and safety operation of the system. In practice, soft faults are very common and have gained a lot of attention. Due to lack of proper fault model and existence of tolerance, the diagnosis of soft fault is a very hard work.At present, in order to guarantee high fault coverage, adding new test points is widely employed for diagnosis of soft faults in analog circuits. Unfortunately, this method requires use of high-performance test equipment and leads to intensive-labor consumption. Moreover, transient interfere, hot noise and other disturbance cause a high aliasing probability, which make diagnosis of soft fault more difficult. To overcome these issues, this thesis focuses on finding fault model, fault signature extraction and anti-aliasing method. Firstly, based on the information theory, a new representation of the linear system is exhibited and the comparison with the traditional form is given. Second, with the help of nonextensive Tsallis entropy, a class of Gaussian-like probability density function(PDF) with explicit form under the variance constraints are obtained, the physical meaning of the parameter in Gaussian-like PDF is also clarified. Lastly, via the minimum cross-entropy method(MCM) and minimum Fisher information method(MFIM), a group of PDF with the explicit form under the mean constraints are derived, and their application to soft fault diagnosis is also presented.The main contributions are:At first, via information viewpoint of linear system and noise-based signal processing technology, a dynamical fault model of analog circuits is set up which combines the measurement equation with the system propogation equation together. Consequently, the fault diagnosis problem is turned into the problem of disturbance estimation of the linear system. Meanwhile, a new use of Kalman filtering method is presented to implement the noise estimation of the linear system. Experimental results show that the proposed method gives a positive consequence in both anti-aliasing and efficiency of the diagnosis.Then, analyze the source of noise and its features, explore the maximum entropy principle and extreme Fisher information method, both of which lead to Gaussian distribution. With the aid of one basic quantity in nonextensive statistical mechanics called Tsallis entropy, a new kind of PDFs named q-Gaussian density are derived. Via q-Gaussian density, an explicit relationship between measurement times and the parameter in q-Gaussian density is established, which implies that increasing the measurement times will not always produce a higher measurement accuracy in the diagnosis process. Meanwhile, this relationship provides a guide to choose the optimal measurent times and this will help reduce the cost of testing. Experimental results prove the correctness of the proposed theory.Thirdly, Illustrate the method for fault diagnosis of analog circuits based on MCM. It proves that the cost of testing and the probability of aliasing will be reduced if the proposed method is employed. Considering Shannon entropy is a special case of cross-entropy, the feasibility of using MCM as a diagnosis method is clarified. From the diagnosis point of view, the tolerance effect can be considerably suppressed via using MCM. Meanwhile, based on AR model, a new algorithm for diagnosis of analog circuit is proposed and the experimental results show that the test efficiency can be improved and the cost and complexity of testing will be reduced if this new algorithm is employed.At last, analyze the method for fault diagnosis of analog circuits based on MFIM. The PDFs with explicit form are derived and it is noted that these PDFs can describe the influence of the tolerance quantitatively in the process of diagnosis. Compared with the maximum entropy principle, it proves that the MFIM can further suppress the effect of tolerance and lead to a more reliable diagnosis result. An experiment is carried out to verify our declarations.