Hydraulic Pump Fault Diagnosis Based On Multi-sensor Data Fusion

Engineering vehic le power transmission and servo control is largely dependent on the core components of a hydraulic pump which is generally in poor working conditions, because of the high strength of its work, the failure mechanism of pump is complex and diverse, different faults may have s imilar status, the parameters of failure characteristic is vague to determine when use a single sensor to detect the failure. Therefore, utilization of multi-source state information in a timely and accurate diagnosis and prediction of their failure has important practical significance.The main contents are as follows :(1)Combined with a hydraulic pump failure characteristics built over three source data fusion fault diagnosis model, the data collection stage acquisition pump casing vibration signal above the hydraulic oil, oil pressure signal in the system and the oil outlet temperature signal leak in the mouth, and then normalized data for each fault; feature level do subnets diagnosis by constructing three separate parallel PSO-BP neural network; decision-making level fus ion diagnostic results of each subnet by using the improved D-S evidence theory, and ultimately achieved an accurate diagnosis of several common hydraulic pump failures.(2)In this paper, diagnosis subnet constructed by BP neural network, however, BP neural network us ing a grad ient descent method in network training process, the network has slow convergence and easy to fall into local minimum value, therefore, combining the PSO algorithm and BP neural network algorithm optimization neural network weights and threshold value matrix can overcome these shortcomings. In the actual binding process, when the dimension of PSO algorithm is large, it may produces a large number of invalid iteration, the convergence rate declined and the search process stalled. Therefore, we made improvements to the PSO algorithm, so that convergence speed and accuracy of the algorithm has been improved.(3) In dec ision-making level, Since the D-S theory cannot be effective synthesis in the condition when evidences are highly conflicting, this artic le deal with the source of evidence, make full use of the correlation between the various evidence, fixed evidence data itself, simulation examp les showed that this method can effectively blend conflict of evidence. Compared with other improved algorithm, this algorithm converges faster and has higher fusion accuracy.(4) Application of three grades data fusion fault diagnosis model in 25SCY14-IB-type axial piston pump, its specific fault experimental analysis, diagnosis achieved good results.