Study Of Knowledge Acquisition Methods In The Tribo-system Condition Identification

Tribo-system condition identification has been developed in the direction of intelligence through nearly 20 years development. The intelligence of tribo-system condition identification mainly reflects on two aspects: one is to estimate wear state of machine's tribo-system according to monitoring information with some mathematical methods; the other is to identify wear state by tribo-system condition identification knowledge which is acquired with knowledge acquisition methods. Knowledge acquisition methods are important for tribo-system condition identification. Owing to the inherent characteristics of tribo-system, the sharing of tribo-system condition identification knowledge is poor. A set of knowledge acquisition methods is needed which possesses the quality of transplant and can apply different monitoring objects to acquire its own tribo-system condition identification knowledge.In this study, sliding bearings are selected as objects. Focusing on the characteristics of tribo-system condition identification, this study firstly studied for theoretical methods to find out the theoretical basis to solve problems, and introduced the elementary theory knowledge. In order to prove the validity of theoretical methods which were used in tribo-system condition identification, the experiments of typical wear stages of sliding bearings were carried out using MMW-1 wear trial machine and internal combustion engine tribology and dynamics trial system. The information of wear particles and wear surface were collected by muti-monitoring equipments.First, the qualitative mapping relationship between wear surface and wear particles was analyzed which came from different typical abrasion process of sliding bearings. Second, the set pair analysis method was used to establish the quantitative mapping relationship from different aspects because of the diversification, incompatibility and inconsistency of descriptive information in tribo-system. The relationship values between wear particles and every typical wear surface were calculated. The biggest value means the unknown condition was closer to this wear state than to others. This method can integrate qualitative and quantitative information. Therefore the identification result was more reliable and the state descriptive information was more comprehensive. Last, tribo-system monitoring information became more and more, and possessed redundancy and relevancy among monitoring attributes, which were unfavorable for automatic identification with machine learning methods. Rough sets and PCA (Principle content analysis) were applied to reduce the amount of attributes. Moreover support vector machine was adopted to find the mapping relationship recognizer between wear particles and wear surface information.Knowledge engineering of artificial intelligence was introduced to tirbo-system. The tribo-system condition is identified with knowledge to enhance diagnosing efficiency, save time and economic cost. According to the characteristics of tribo-system and the concept of knowledge acquisition, tribo-system condition identification knowledge can be obtained from three aspects: experiential knowledge, trial data and monitoring cases. The acquisition of experiential knowledge in tribo-system condition identification which is acquired by filtering, abstracting and summarizing from former monitoring information is one of the main sources of tribo-system condition identification knowledge. Aiming at trial data whose information quantity is much bigger, the knowledge acquisition model based on Bayesian networks was established and the notional and disciplinary condition identification knowledge was acquired which was significant for tribo-system condition identification. Considering the facts that there are some machines which are not used for a long time and the monitoring cases are not so many, decision tree was introduced to acquire tribo-system condition identification based on monitoring cases. It can abstract simple and practical tribo-system condition identification knowledge.