Study On Soft Fault Diagnosis For Analog Circuits Using Sensitivity Vector As The Fault Signature

The field of analog circuits has undergone remarkable changes over the past decades. Fault diagnosis and testing of analog circuits have grown into a special field of interest in semiconductor industry. The methodology for testing of digital circuits is well established. However the methodology for fault diagnosis of analog circuits is relatively unexplored. Some of the reasons for this under development are lack of proper fault model, continuous variation of the element parameter, parameter tolerance and measurement error, etc. Based on circuit theory, matrix theory, optimization techniques and signal processing techniques, etc, soft fault diagnosis for analog circuits using sensitivity vector as the fault signature was further studied. The research results are as follows.It is proved that the direction of the nodes increment voltage vector is determined only by the location of the faulty element, and it is independent of the parameter deviation of the faulty element. Methodology of fault diagnosis using sensitivity vector as the fault signature is proposed.The bound of the current deviation at each port is used as the weight to establish the fault pattern matrix from the sensitivity vectors.The fault diagnosis equation is then established by the fault pattern matrix and the nodes increment voltage vector. Then the problem of fault diagnosis is formulated into finding the solution of the fault diagnosis equations.The character of the optimal approximation is discussed including the minimal norm character and the least square character. The method of fault diagnosis by using Moore-Penrose inverse to solve the fault diagnosis equation is proposed.The validity of the method is proved by the simulation results.The adverse impact to fault diagnosis from the redundant test nodes is discussed, and a method for eliminate the redundant test nodes based on fuzzy clustering is proposed. It is also extendedly used in defining the fault sets of the circuit.It is proved that the condition number of a matrix is bigger than the maximum ratio of the norms of the row vectors.Then a method of nodes selection is proposed considering both the relativity between the row vectors and the condition number of the matrix selected.The problem of analog circuits fault diagnosis is formulated into the problem of multiple objective linear programming. A multiple objective linear programming method, say two-step method, is proposed to localize the faulty element. Considering the computation complexity of the simplex method, a fault diagnosis methodology based on primal-dual interior point method is proposed. Thus the faulty element can be localized by less computational cost.The process of the circuit fault is discussed, and then a fault diagnosis methodology based on online monitoring is introduced. The fault diagnosis methods based on modified AMUSE algorithm then is proposed.