Method Of Knowledge Learning For Fault Diagnosis Based On Inductive Learning Strategy

In the process of machine reasoning, the help of a priori domain knowledge is needed. In many cases, there are areas of incomplete knowledge of the existence domain, as follows:(1) Existing prior knowledge and domain knowledge obtained through training by a small amount of sample (partially covered), can not fully cover the whole sample of the general distribution, which is regarded as approximate priori knowledge.(2)The existing approximate domain knowledge may include errors.Under the guidance of such incomplete and even error-containing knowledge structure. Generalization accuracy, based on the deduction of knowledge, drops and even leads to erroneous or false judgments. Therefore, increasingly grasping the knowledge through accumulation of samples and gradually forming a more complete knowledge of the field, is an effective way to improve the generalization ability of intelligent reasoning system.This article puts forward a fault diagnosis of knowledge learning method which is to establish the initial field of knowledge based on inductive learning knowledge to areas of incomplete knowledge and a small amount of training examples as a starting point; at the same time in the production process, with the accumulation of samples, establish new general assumptions consistent with domain knowledge and accumulated data, through new knowledge gained during the latter part of the learning process. This model, under the guidance of domain knowledge, spawns knowledge acquisition and knowledge learning training, thus ensuring the strong control of existing good knowledge in the process of knowledge discovery and inductive generalization which provides a powerful solution whose early period is dominated by domain knowledge while gradually change into inductive machine learning during the middle and late period.As for this article, firstly a decision-tree-based inference engine model and earning of inference engine are proposed. Moreover, initial field of knowledge is established based on priori domain knowledge, small amount of sample data and then the establishment of the initial assumptions, and decision tree learning and training. As the constraints of actual engineering problems to domain knowledge leads to insufficient sample coverage, it is difficult to form a complete a priori knowledge system, and thus confine the process of logical reasoning based on knowledge, which directly reduce the generalization accuracy of inference engine based on knowledge.For this problem, the article proposes a knowledge learning model,automatically get new knowledge based on Genetic search algorithm, under the guidance of priori domain knowledge. Fitness function and evolution strategy are the core of genetic algorithms. This model is to change the key data search genetic algorithms of global greedy search to new general assumptions whose process is to get new knowledge gradually and reject incorrect prior knowledge and deduce a highly general assumption consistent with the domain knowledge and data. In this model, the introduction, guided search and evolution of domain knowledge is within the control of existing excellent knowledge guide therefore new areas of knowledge with a higher generalization accuracy is able to achieve.On the basis of the above and combining the experimental field data collection of automobile transmission, this paper establishes a unified decision-tree inference engine. Through the unit as well as joint testing and generalization accuracy review of decision-tree inference engine in the early period and the inference engine set up in the later period, the results show that this model is able to establish the initial general assumptions based on similar field theory (incomplete priori knowledge) and a small number of training sample (partially covered) to achieve learning and training; also has self-adaptive learning ability and support the learning process in the latter part of the discovery of new knowledge, supplement and amend the domain knowledge, reasoning out the new general assumption consistent with new and old data, and then automatically make adjustment to the reasoning model in the level of machine model...