As a key functional component of the processing centre, the cylindrical bank has complex structure, speed of change and high accuracy of location, making it a highly probable component of failure in the processing centre, which seriously affects the reliability of the operation of the processing centre. Owing to the variety of failure patterns and the multiplicity of reasons for failure, it is difficult for maintenance staff to identify the cause of failure and quickly repair it in the event of failure。
The main research and innovation points for this paper are the following: (1) the sensitive signals of the operational state of the knife are identified. Based on the on-site reliability test data of the processing centre, a common failure pattern of the processing centre's twilight bank system was analysed and, based on a failure factor analysis, sensitive signals were identified that reflected the normal and malfunctioning state of the knife bank (see para. Vibration signals) are also collected using the status monitoring system of the reliability test station of a disk bank for sensitive signals indicating the operational status of the knife bank。
A cs-svm-based cylindrical knife bank failure sorter was established. In the first instance, the vibration signals collected by the test-platform monitoring system were characterizationd and the post-exposure characterization was normalized, resulting in an initial set of features that reflected the operational state of the disk bank. On this basis, a cs-svm-based cs-turkish blade fault catalog was established using the cuckoo seach, cs, optimizing support vector machine, svm。
Reductions in the initial feature set have been achieved by ensuring the accuracy of the classification. First, a 5-storey small-wave package decomposition of the vibration signals obtained 32 small-wave coefficients, and the energy values of these coefficients, which were integrated, were used as the initial characterization set; the importance of a concentration of the initial characteristics is then evaluated, taking into account the successor classification; finally, the taxonomy was trained and tested using a forward-sequencing search algorithm and the accuracy of the taxonomy was used as an indicator for evaluation, and the feature subsets that enabled the taxonomy to achieve the best performance and contained the minimum number of fault features were selected as the most effective failure feature subsets based on the results of the evaluation indicators. This allows the classificationer to be trained and tested directly, using the characteristics corresponding to this optimal subset, both for optimal performance and for significantly reduced computing complexity。
