Application And Research Of Wavelet Neural Network In The Pot Vessel Defect Detection

As pot vessels are getting heavily used, more and more attentions have been paid to the security and defect detection methods. Therefore, there are important realistic meaning and useful value to study a fast, high-performance, portable defect detection system, which is easy for data management and can be used in real-time monitoring.Currently, the fault diagnosis of acoustic emission technique in pot vessel used 'wavelet packet-energy' as the input feature extraction of neural network usually. This method is capable of decomposing high-frequency and low-frequency signals, but the disadvantage is that the decomposition is not self-selective by the band's characteristics. Since the analysis capacity of wavelet packet in time-domain is not fully appreciated in this method, by taking this into consideration, it will help to improve the extraction of the characteristics of defects signals.In view of above questions, a featured extraction method based on 'wavelet packet of sections and energy-moment' is proposed. It can decompose the signals adaptively according to the energy focus key band by taking advantage of the wavelet packet time-domain information. At the same time, in the light of slow convergence, long time iterative and other issues of traditional BP neural network, a gradient algorithm is used to optimize the network and an improved activation transfer function is used for BP algorithm. According to the result, this system has greatly reduced the complexity of the algorithm and increases the recognition precision. Meanwhile the defect locations are measured by a single sensor, this system has shown to have great value of popularization.To meet the needs of high speed sampling and high data managing capability, the system adopts double-CPU of TMS320C6713 (DSP) and MSP430F149 (MCU) in the hardware design. DSP is used for signal acquisition and processing, such as wavelet, neural network algorithm and so on, MCU is responsible for managing man-machine peripherals, data communication and I/O controlling.