The Research Of Data Driven Fault Diagnosis Based On Residual Evaluation

Fault detection and isolation (FDI) system is the main component of intelligent facility, plant and service platform, and FDI is also a key technique for Industry 4.0. Faults are divided significant fault and incipient fault from a new perspective in this paper. Incipient fault is forewarning of significant fault, and incipient fault diagnosis can avoid accidents or reduce the costing. Thus, this work presents several FDI approaches for two types of fault based on residual evaluation and data driven method. Furthermore, a few of novel FDI solutions are proposed for the complex industrial process.Firstly, a new fault isolation strategy is presented based on residual reconstruction and contribution plot analysis. Based on the space projection, residual evaluation and contribution plot is unified into a framework. In this paper, parity space and subspace identification methods are used to generate residuals for fault detection. Then the optimal residuals are utilized to obtain process fault isolation scheme. A new contribution index is calculated according to the average value of current and previous residuals. Simulation results show that smearing effect can be eliminated and fault evolution can be acquired based on this index. Moreover, this strategy extends the usable scope of residual evaluation.Secondly, incipient fault detection is studied based on feature extraction, and this can probe the fault information beforehand. Based on a novel three stages of process variables, a two-step incipient fault detection strategy was proposed for monitoring the complex industrial process. The first step aims at the significant fault detection using the traditional multivariate statistical process monitoring methods. Then a method combined the wavelet analysis with the residual evaluation was carried out for monitoring the incipient fault. Wavelet analysis aims at extracting the incipient fault features from process noise. The residual generation is optimization based on the robustness and sensitivity index (RSI), which can be realized directly using the test data. An improved kernel density estimation based on signal noise ratio is proposed to adaptively determine the detection threshold. The proposed incipient fault detection scheme is tested on a numerical example and the Tennessee Eastman process, and better monitoring performances are obtained comparing to other traditional fault detection method.Finally, we study the incipient fault based on the parameter variation, and the main technique is to design the auxiliary signal. Active fault detection (AFD) based on residual ellipsoid is presented and the novel auxiliary signal and fault detection logic are proposed. Firstly, two new Fitness Functions are used to design the suboptimal auxiliary signal based on residual ellipsoid. This can guarantee that a better trade-off between the higher fault detection rate and the lesser impact on system. Secondly, novel fault decision logic for AFD is obtained. The logic is based on the end-residual of test period and the real-time tracking residual. The end-residual is the main evidence for decision making, and the tracking residual is utilized to probe the abnormal condition of the process beforehand. These two methods are complementary to each other. Finally, a numerical case is utilized to demonstrate the effectiveness about the proposed approach.