Fault Diagnosis For Analog Circuits Based On AdaBoost Algorithm

The research and design of the marine fuel flowmeter is an important project derived from "the study on the energy saving technology for ships sailing in Three Goreges Reservoir" developed by Chongqing Commnucations Committee.Due to its growing complexity, analog circuits in modern electronic systems are suffering high risk of faults and pose great difficulty in the fault diagnosis for analog circuits. Any fault of analog circuits may affect the accuracy and stability of marine instruments and subsequently cause damages to marine instruments. To increase the accuracy and stability of the marine fuel flowmeter, the study on the fault diagnosis for analog circuits shall be made while the research and design of the marine fuel flowmeter is carried out. The self-diagnosis function may be rendered to the marine fuel flowmeter without affecting the inherent functions of the marine fuel flowmeter.Infinite AdaBoost algorithm can be used to carry out the fault diagnosis for analog circuits. In the course of pattern recognition, it is proven that it is difficult to find strong classifiers with high classification accuracy and easy to find weak classifiers with classification accuracy just better than random guessing. By combining weak classifiers, AdaBoost algorithm can create a strong classifier so that the strong classifier for fault diagnosis for analog circuits needn't be searched directly. In order to increase the classification accuracy of the AdaBoost algorithm, infinite hypotheses are combined into the AdaBoost algorithm for the infinite AdaBoost algorithm. Then, fault diagnosis for analog circuits can be carried out through infinite AdaBoost algorithm.This thesis introduces infinite AdaBoost algorithm based on the Support Vector Machine, and analyses associations between the AdaBoost algorithm and the Support Vector Machine.The associations are as follows.(1) According to the Lp-norm maxim margin theory which is used to analysis the optimum goal for weights of the AdaBoost algorithm, the optimum goal of the AdaBoost algorithm is perfectly equivalent to that of the Support Vector Machine.(2) According to the similarity analysis between the classifier of the AdaBoost algorithm and the classifier of the Support Vector Machine, the classifier of the AdaBoost algorithm is perfectly equivalent to that of the Support Vector Machine by predetermining conditions.(3) According to the similarity analysis between the mapping componentφ(x) of the Support Vector Machine and the weak classifier h (x) of the AdaBoost algorithm, the mapping componentφ(x) of the Support Vector Machine is perfectly equivalent to the weak classifier h(x) of the AdaBoost algorithm.Due to the predetermined associations between the AdaBoost algorithm and the Support Vector Machine, the infinite AdaBoost algorithm can be realized by the Support Vector Machine.The key of the framework is to embed an infinite number of hypotheses into a Support Vector Machine kernel.The infinite AdaBoost algorithm was programmed by using Matlab 6.5 and the faults of analog circuits were diagnosed by means of the infinite AdaBoost algorithm. The result of the fault diagnosis for analog circuits indicates that the classification accuracy of the infinite AdaBoost algorithm is better than that of the finite AdaBoost algorithm and the classification accuracy of the AdaBoost algorithm is improved by using the infinite AdaBoost algorithm.The study on the fault diagnosis for analog circuits was made while the research and design of the marine fuel flowmeter was carried out. The research and design of the marine fuel flowmeter was completed and on-board experiments for the flow rate were completed while writing this thesis. During the process of designing the marine fuel flowmeter, the stability and reliability of the marine flowmeter was analyzed and was improved by means of the fault diagnosis for analog circuits. Self-diagnosis function was rendered to the marine fuel flowmeter without affecting its inherent functions.