Engineering ECU Detection System Of Vehicle Engine Based On Virtual Instrument

The Electronic Control Unit(ECU)of engineering vehicle engine is responsible for controlling the coordinated operation of all parts of the vehicle,and its health status directly determines the working efficiency and safety performance of the engine.Due to the bad working environment,complicated internal circuit and other factors,engine ECU failures occur frequently.Therefore,research on ECU fault detection is of great significance to ensure the safety of vehicle operation.Based on engine ECU as the research object,this paper aimed at the present stage the signal producing technology is limited by the hardware simulation and difficult to achieve high precision ECU fault is difficult to detect the key issues of research and discussion,and combined with virtual instrument technology and deep learning technology,to carry out the signal generator design and research of injector drive circuit fault diagnosis method.In this paper,two key technologies involved in ECU detection are studied as follows.(1)Aiming at the problems of low accuracy and poor flexibility of the simulation signals simulated by existing signal generators,a segmentation model of an ECU excitation signal based on characteristic parameters is proposed.According to the basic structure and working principle of the sensor,the whole signal is divided into several local signals,and the local signals are simulated by means of mathematical modeling combined with the characteristic parameters of each part.By adopting the same global parameters and strictly controlling the proportional coefficients between different parts,the synchronous output and dynamic frequency modulation between multi-channel signals can be realized.Simulation experiments and spectrum analysis show that the simulated signal retains the effective information of the original signal.According to Pearson similarity theory,the similarity degree between the simulated signal and the actual signal of HBS-6 integrated detector is 48%,and the similarity degree between the waveform signal of crankshaft missing tooth simulated by arcsine function and the actual signal is 68%.However,the similarity between the simulated signal generated by the piecewise modeling method proposed in this paper and the actual signal has reached 74%,which has a strong correlation,and the accuracy of the simulated signal has been improved compared with the above two methods.(2)A fault diagnosis method for injector drive circuit based on one-dimensional convolutional neural network(1D-CNN)is proposed to solve the problems of obscure fault characteristics and difficult detection of ECU.The data set of training network is extended by simulating the fault of fuel injector drive circuit offline,and the characteristic information of fault waveform is extracted by convolution kernel.Through model comparison experiments,the average detection accuracy of LSTM is 71.78%,the average detection accuracy of GRU is 72.43%,and the average detection accuracy of RANDOM forest RF is 97.20%.However,the 1D-CNN model method has higher identification accuracy,and the average detection accuracy of various types of injector fault data sets can reach 97.56%.Compared with the above model method,the average detection accuracy of this data set has been improved to a certain extent.(3)According to the above two methods,this paper developed the engineering vehicle engine ECU detection system based on virtual instrument,and completed the prototype of ECU comprehensive detection device.The detection system mainly includes ECU excitation signal generator module and ECU fault detection module.Data transmission between engine ECU and detection device is realized by using data acquisition card USB3151,and the visual development of related functional interface of onedimensional convolutional neural network is completed.The experiment shows that the ECU excitation signal simulated by the signal generator module of the prototype device can start normally and affect the work of ECU,and users can intuitively and clearly display the fault detection results of 1D-CNN network.