| 摘要 |
<p>PURPOSE:To diagnose a failure without preparing a sensor for failure diagnosis by extracting feature variables from the I/O data of a control device and comparing the extracted feature variables with a previously set normal feature variable distribution range. CONSTITUTION:Input/output(I/O) data, i.e., the data of a control variable (y) and a manipulated variable (u), are stored in a data storing part 10 and plural feature variables are extracted from these stored data. The distribution of respective extracted feature variables is found out, and when the distribution has no dispersion, the distribution is decided as a normal state and a normal feature variable distribution range is set up. A decision part 12 compares respective feature variables extracted by a feature variable extracting means 8 with the normal distribution range set up by a setting part 11. The abnormality of I/O data is decided by combining combined results and a corresponding abnormal signal is outputted as a decided output. Thus failure diagnosis can be attained without preparing a sensor for failure diagnosis.</p> |
