Fault Diagnosis Approach Based On Neural Network To Chemical Process

Chemical process monitoring and fault diagnosis ,which are the intensive research subjects in modern chemical process, assure the stability and security of production process. Fault diagnosis of the chemical process was discussed and the thesis proposed that a deformation of radial basis function neural network– probabilistic neural networks(PNN) were applied to fault detection and diagnosis(FDD) of the chemical process. FDD ability was validated by a example and Tennessee Eastman(TE) process . The results were compared with the BP networks,and showed that PNN gave a better performance than the BP networks on FDD.On the basis ,The thesis proposed the method that integrate Haar wavelets and PNN(HWPNN) for fault diagnosis. TE process was used to validate the effectiveness of HWPNN. The simulation results showed that the fault diagnosis accuracy of HWPNN reach 100%. So HWPNN is feasible and effective in the fault diagnosis.The main contribution can be summarized as follow:1. FDD methods were reviewed, and the development of FDD were analyzed.2. Tennessee Eastman process is introduced to validate the effectiveness of FDD methods.3. BP networks were studied. BP networks were applied to FDD and was validated by a example and TE process .4. The thesis proposed that PNN was applied to FDD of chemical process. Then TE process were used to validate the fault diagnosis ablity of PNN .5. The thesis proposed the method that integrated Haar wavelets and PNN(HWPNN) for fault diagnosis. And HWPNN was validated by taking TE process as example. The sumilation results showed that HWPNN is feasible and effective on fault diagnosis of chemical process.