Research On Soft Fault Diagnosis And Parameter Identification Of DC-DC Converter

As the key component of switching power supply,DC-DC converter(Direct current-Direct current converter)is widely used in many important fields such as national defense and aviation,etc..The fault states will directly affect the reliability and stability of the entire power supply system.In the actual operation of the DC-DC converter,the parameters of the internal components will deviate from the normal values under multiple environmental stresses,then the soft faults will be caused.In the early stage,the degree of the parameter deviation is small,the fault features are weak.As the working time increases,the deviation degree will gradually increase.When the component parameters exceed the failure threshold,which will cause serious fault of the DC-DC converter.It will result in the anomalous function or even breakdown of the entire power supply system,and then result in the shutdown of the entire equipment.Therefore,the paper focuses on the soft fault caused by the parameter deviation before the deviation exceeding the failure threshold,and proposes the effective fault detection,diagnosis and parameter identification methods for DCDC converter.It will provide the strong technical support for condition-based maintenance and health management of the power system.The paper researches the detection,diagnosis and parameter identification for the soft faults in DC-DC converter.The main research work are as followed:(1)Aiming at the problems including the difficulty of determining the detection threshold and the unstatisfied soft fault detection effectiveness caused by the weak fault features for DC-DC converter,a soft fault detection method based on Gaussian Process Regression is proposed.Gaussian Process Regression is used to reasonably estimate the normal output range,and the estimated range can be automatically adjusted based on the fluctuation of the output signal and the noise.To reduce the effect of the local interference on the soft fault detection effectiveness,the norml output range will be expressed as the extreme values of multiple statistical features.The extreme values are used as the important criterion for soft fault detection.The results in simulation and hardware experiments demonstrate that the proposed method can cover more fault types including hard faults and soft faults caused by single and multiple components.The proposed method also has stronger fault detection ability,especially for the soft fault caused by the weak parameter deviation.It demonstrates that the soft fault detection threshold is reasonable and practical.(2)Aiming at the problems of low diagnosis accuracy for soft fault in DC-DC converters and high dependence of diagnosis models on fault samples.A soft fault diagnosis method is proposed based on overlap evaluation.Firstly,for the key soft fault types caused by weak parameter deviations of single and multiple components,the overlap evaluation is proposed to select the sensitive features,which increase the distance between normal samples and fault samples to improve the soft fault diganosis accuracy.Secondly,in order to solve the problem of inaccurate fault diganosis model caused by the lack of fault samples,support vector data description(SVDD)is used to construct the diagnosis model for each key soft fault type.Finally,all diagnosis models are executed in the order of the occurrence probability of the key soft fault types.The diagnosis models can obtain great diagnosis accuracy for different soft faults.The experimental results are as followed.The advantage of the proposed method is selecting sensitive features.The fault diagnosis models based on SVDD can obtain great soft fault diagnosis effectiveness.The diagnosis method has low dependence on fault samples and high fault diagnosis accuracy,it demonstrates that the method has great practicability.(3)Aiming at the problems of the unsatisfied identification effectiveness for the weak parameter deviations in DC-DC converters and the low parameter identification accuracy under fluctuating input voltages,a parameter identification method based on Dendritic network is proposed.For different parameters,the coefficient of variation evaluation is proposed to select the key features,which are combined with Dendritic network to establish the multiple component parameters identification networks to improve the identification accuracy.Considering that the effect of fluctuating input voltages on the parameter identification accuracy,the input voltage identification network is established based on Dendritic network.Then multiple component parameters identification networks under fluctuating input voltages are established by combining the input voltage identification with parameters identification networks.The experimental results demonstrate that the proposed identification networks can obtain higher parameter identification accuracy for the weak parameter deviations and fluctuating input voltages.Compared with the existing parameter identification methods based on artificial intelligence,the established parameter identification networks based on Dendritic network have stronger generalization ability and analysis ability of data hidden laws.The proposed method effectively improves the parameter identification accuracy,and has great feasibility and applicability.(4)In order to comprehensively verify the soft fault detection,diagnosis and parameter identification methods proposed in the paper,the engineering validation is established based on the power supply circuit of controlling board.Firstly,all kinds of soft faults are injected into DC-DC converters,and then the soft fault detection,location and parameter identification are implemented separately to DC-DC converters.The experimental results show that the methods in this paper can effectively recognize all kinds of soft faults of DC-DC converters.It demonstrates that the proposed methods have great practical value.