Study Of Fault Detection Based On Process Data

Improving product quality and reducing process fault are two important issues that industries have been paying great attention to. It is the key target to keep the process operating under "Safe, Stable, Long-term, Full-capacity". Process fault detection is one of the most effective technologies to achieve the target, And process fault detection is very important to improve the efficiency and profits too.In this dissertation, the local tangent space alignment combined with the independent component analysis(LTSA-ICA) and the local tangent space alignment combined with the support vector data description(LTSA-SVDD) methods are proposed to cope with the process with nonlinearity, high-dimension and complex distribution. Then they were tested on Tennessee Eastman Process(TE) and successfully detected some faults. And the performances of the two methods were compared by simulations. A batch process fault detecting approach based on multiway factor analysis combined with the independent component analysis (MFA-ICA) was proposed. It was used to deal with the typical batch production process with low-dimension and huge samples. This method was tested in penicillin fermentation process. Furthermore, multiscale factor analysis method(MSFA) is proposed to improve the precision of fault detection in multi-time-frequency space. The performance of the method was illustrated by a numerical example. At last, fault detection software been developing was introduced.