Study On Blind Source Separation Based On The Application Of Mechanical Fault Diagnosis

Fault diagnosis techniques for mechanical devices are of great significance to the devices’ safe and efficient operation. Among them, the method of using noises produced by mechanical devices to diagnose the faults has already become an important direction in the field of mechanical fault diagnosis. When mechanical devices operate abnormally, various noise signals are usually produced. Mechanical faults can be diagnosed accurately through signal processing methods by separating various noise signals and transforming them into independent signal sources to extract fault information. Blind Source Separation(BSS) techniques can identify and separate the source signals under the condition of noisy signals and mixed transmission of the signal channels, thus they provide solutions to the test of the device’s status and the diagnosis of the faults in noise environments. Based on the two common categories of blind source separation: determined BSS and underdetermined BSS, this paper presents respectively the linear BSS algorithm of the neural network and the underdetermined BSS algorithm of the least square method and verifies the validity of the fault diagnosis on the mechanical system. The main research includes:(1) Linear BSS algorithm of the neural networkThe paper presents two BSS algorithms of self-adaptive linear neural network(SALNN-BSS) and cerebellar model articulation controller(CMAC-BSS); builds separately their structure models and solution algorithms and compares the separating effects of SALNN-BSS and CMAC-BSS by comparing the function coefficient of emulational mixed signals. By doing this, this paper finds that both SALNN-BSS and CMAC-BSS can successfully separate the mixed signals, and the recoverable source signals have good waveforms, but CMAC-BSS has smaller separation errors and can separate the signals more swiftly and with smaller stabilization errors, thus it is more suitable to be used to identify BSS faults.(2) The underdetermined BSS algorithm of the least square methodCombining the characteristics of the least square method and the underdetermined BSS, the paper presents the underdetermined BSS algorithm of the least square method, formulates the operating steps of the algorithm, testifies theoretically the uniqueness of the algorithm’s solution by deduction, adopts the algorithm to separate respectively the mixed overlapping voice signals of sparse and nonsparse and builds their mixed models and mixed matrix. By comparing the signal-to-noise ratio of the source signals and the separated signals, the paper proves that the algorithm is very effective no matter in separating sparse voice signals or nonsparse voice signals.(3) The application of BSS techniques in fault signals diagnosis of ice-cooled storage cabinetsThe paper builds the experiment system of the fault diagnosis of the ice-cooled storage cabinets based on the principles and the commons faults of cabinets, and completes the circuit design and the structure designs of the touch panel operation interface, etc. The system can be used to diagnose the failures of the axial flow fan of cabinets and the failures happening in axial flow fan and water pump at the same time. By comparing the ability of SALNN-BSS and CMAC-BSS to separate the signals of corresponding faults, the paper finds that CMAC-BSS is more precise in recovering original signals, therefore, it has better recognition effect and is better in immediacy in identifying the faults.(4) The application of BSS techniques in the diagnosis of the engine’s noisy signalsThe paper builds the experiment platform of the noise test of the engine and detects the mixed voice signals with the engine operating. By adopting ME-BSS and CMAC-BSS to separate respectively the interference noises produced by the engine and the outside world, the paper proves that CMAC-BSS has better separation effects than ME-BSS, and can separate quite well the engine’s noises from the external noises. CMAC-BSS has been further applied to diagnose the engine’s abnormal faults and the experiment shows that CMAC-BSS can capture the fault signals in the engine’s movement process.(5) The application of underdetermined BSS techniques in the fault diagnosis of capsule CO2 air conditioningThe paper analyzes the characteristics of common faults signals(vibrating signals and noisy signals) of CO2 air conditioning system which is the most important part of the rescue cabin, designs and builds the testing platform of the fault signals, separates the systematical fault signals with known and unknown fault numbers respectively by adopting LSM-UBSS and probes its separating errors. The paper finds that LSM-UBSS can be used to diagnose systematical faults of CO2 air conditioning with the known fault numbers. With the unknown fault signal numbers, the number of the fault source signals can be effectively identified and the source signals can be recovered by adopting first the subtractive clustering BSS algorithm to estimate the mixed metric vector, and then LSM-UBSS algorithm.