Research Of BP Neural Network In Solving Parameter Selection In Automatic Manufacturing

The parameter selection in Automatic Manufacturing is the bottle-neck ofCAPP(Computer Aided Process Planning) for a long time.To solve this problem,weused ANN as a new method and took reflection phenomenon in Manufacturing asan example to conquer this question.The reflection phenomenon in the processing of shaft type parts influnence theprocessing precision. As the factors of affecting the processing precision which aredifficult to define are so many, we can't find a formula for calculating to know howto process the part to attain the expected processing precision. Even though we canfind the formula, it is a trouble to work out the result.The emergence of neural net arithmetic solves many problems that can't besettled by formula, Through the training of swatch collections and learning thestatistic rules, NN stores the knowledge learnt after training in weights. When thepatterns that not included in swatch collections the true value. BP(back propagation)arithmetic has very strong non-linear mapping capacity, and is not limited by theinput numbers and output numbers. In the actual research, you can modify theprogramme freely as you need. BP arithmetic is being applied widely nowadays,and the improved arithmetic based on BP often comes forth. Aiming at the actualproblem this paper researches, I use BP as the researching tool. BY using plenty ofexperiment data to train the BP, I research and analyze the factors influencingreflection problem and the probability of using BP to solve the reflection problemin processing the shaft parts.The conresearching contents are as follows:1. The factor's influence on the processing processing precisionWe find that some input has very important effect on the result in theexperiment ,which reminds us that this input is the decisive factor influencing theprocessing precision. But some input has little effect on the result ,then we canconclude that this input is not decisive factor and it can be neglected.2. The confirmation of the net structureThrough many experiments, this paper validates and summarizes how toascertain the structure of the BP net and the different net structure's effect onexperimental results. The confirmation of hidden layers and the nerve cell numbersof a hidden layer mainly.3. The research of BP net in solving parameter selectionAfter the training, we use the test collections to test the reasoning capacity ofBP net. Analyzing the feasibility of using BP net to solve the reflection problem byobserving the difference between the test results and the actual results.4. The part modification to the BP arithmetic progammeIn various researching tasks, something it is necessary to modify the BPprogram partly to partly to accommodate the need of the task. In this paper, BPprogramme is modified partly and it runs well.