Research Of Sub-healthy Recognition Algorithm Based On The Improved ELM

With advances in technology, machinery fault diagnosis technology has become increasingly important. In the production line, if one part or one device fails, but there is no time to find it or fix it. This will result in the stagnation of the entire production line of life or cause harm to the staff there. At the same time it has brought huge economic losses to the enterprise and society. According to the research, most of the equipment failure is a gradual process. In this process, the evolution of the status of the device which is called sub-healthy. It can be appreciated that the production equipment in the sub-healthy state is very dangerous. To a certain extent, reduce equipment failure or prolong the life of equipment, research on the sub-healthy state of the device is the urgent problem of industrial production.This article will identify the fan failure of sub-healthy state as the main research content, and start from the vibration of the fan. First, data collection and data preprocessing; Experiments imitating fan blade fracture, rotor unbalance, bearing fault feature loose, fan imbalance, and extracts the most appropriate fault features. The vibration signal acquisition equipment is used for wind turbine testing and diagnosis. Extracting vibration signals, time-domain characteristics of the experimental data used in this paper. Since the data contains a lot of interference information, this paper presents an improved genetic neural network feature reduction algorithm for data dimensionality reduction. The algorithm combines the advantages of genetic algorithm and the neural network algorithm integration, and improved their fitness function. Reduce the computational algorithm recognition, and improve recognition accuracy of the algorithm. Secondly, building sub-healthy recognition model. This paper presents an Improved Fruit Fly Optimization Algorithm to optimize Extreme Learning Machine sub-healthy recognition model (IFOA-ELM). The input weights and hidden layer bias of ELM is randomly selected. This will result in insufficient generalization ability, and therefore the paper presents an Improved Fruit Fly Optimization Algorithm(IFOA) to optimize the parameters. The IFOA added their own learning ability and improved the inertia weight update method. It also joined the optimization within species groups to achieve fine search, and enhance its ability of local search.Here using an acceleration sensor, data acquisition cards, fans and other equipment collected vibration signal. And using MATLAB software to do the simulation. Experiments show that, IFOA-ELM model between the global and local search achieves a good balance. State identification precision and efficiency has been improved.