Study On Change Detection And Application In Boiling Detection

Boiling detection with the detector based on the change of temperature rate can be transformed into the on-line detection for mean change essentially. Detection of abrupt change is the general interest in the fields of quality control, failure detection, etc. However, some necessary steps to cany out the theory of change detection, such as statistical performance analysis, parameter optimization and so on, are tedious and complicated, which lower the usefulness of change detection. Based on the characteristics of boiling detection, the paper put efforts to make the theory applicable, and presents the adaptive algorithm of change detection to monitor the boiling.The model and the noise of heating process are analyzed based on the measured data and the concerned characteristics of the change detection are proposed firstly. Statistical performance analysis and parameter optimization are discussed for the three most frequently used control charts (i.e., Shewhart, CUSUM and EWMA). On the aspect of statistical performance analysis, the evaluation is easy for Shewhart control chart due to the independence of its test statistic, but is tedious for EWMA and CUSUM control charts and has to recur to the numerical methods. Markov chain approach is introducsd and modified so as to be usable for one-sided EWMA control chart. The integral-equation approach is developed as a unified method for run-length statistical analysis. Run-length numerical tables for standard detecting process are calculated by integral-equation approach and optimal parameters corresponding to signal-to-noise ratios are also obtained based on the tables, which simplifies the difficult parameter optimization significantly and makes it possible to adjust parameters on line. In the end, combining with outlier rule, heating process parameter estimation and some measures to ensure computable, change detection runs in embedded system successfully.The methodology outlined above shows good quality when applied in boiling detection, such as good adaptability, little experiment cost, and better detecting performance than expected, which shows the practicability and validity of the change detection theory, and offers a valuable reference solution when applying the algorithm in other cases.