Fault Diagnosis Of Single-chip Switching Power Supply Based On Neural Network

Single-chip switching power supply has many advantages such as a simple circuit and good cost performance, meeting the design requirement of high efficiency and energy saving for industry, applying to a number of practical fields. Due to the lack of using intelligent fault diagnosis technology, it often needs to check components individually with multimeters relying on work experience when diagnosing the fault of single-chip switching power supply, and it means low diagnostic efficiency and online real-time fault diagnostic is unable to be achieved. Moreover, because the tolerance of devices is obvious and modeling of the control chip is inconvenient, it is difficult to apply traditional intelligent fault diagnosis technology.Because of the characteristics such as fault-tolerance, self-adapting and nonlinear, neural network is suitable for analog circuit fault diagnosis in particular. To improve the diagnostic efficiency, a method of fault diagnosis based on neural network for single-chip switching power supply is proposed in this paper, utilizing three layer BP neural network as diagnostic tools. The data of fault feature vector is obtained by collecting the actual output voltages of the power, and the training samples are extended with virtual samples generated by data perturbation. Hardware and software systems are designed and developed, including a multichannel data collecting system based on MSP430and a program for data displaying on the host computer and fault diagnosis. The feasibility of the method and the operating results of the systems are tested with diagnostic experiments of two practical single-chip switching power supplies.The experimental subjects include single fault, multiple faults and hard fault, soft fault. The results of experiments prove that the fault types can be diagnosed accurately online real-time by the data collection system and the program of fault diagnosis. It illustrates that neural network can be applied to the fault diagnosis of single-chip power supply and the method can overcome the devices tolerance, nonlinear and other difficulties, suiting for multiple types of fault. The method with hard and soft systems can meet industry needs.