Research And Development Of Quality Control Chart Pattern Recognition Based On LSTM

Saving resources and optimizing resource utilization are essential parts of sustainable development and important parts of intelligent green manufacturing.The cutting stock problem refers to the reasonable geometric combination of small parts on large materials and determining the layout to maximize the utilization rate of materials.Therefore,the cutting stock problem research is a problem with important economic significance and social benefits.The cutting stock problem has high time and space complexity and is a non-deterministic polynomial problem which is known as NP-hard problem.Generally,there is no polynomial algorithm.Many experts and scholars engaged in related research and application work in recent years and achieved many valuable results.Most experts and scholars’ research focus on solving algorithms,looking for practical search algorithms to search the approximate optimal solution from the huge layout solution space to obtain a higher utilization rate of materials.In this thesis,one-dimensional cutting stock problem and two-dimensional guillotine cutting stock problem in small batch and multi-batch production mode are studied.It is proposed to predict the parts’ demand of subsequent batches by using Long Short-Term Memory network(LSTM).According to the idea of centralized cutting,the predicted parts’ demand of multiple-batches is integrated into a large-scale cutting stock problem,and the problem of parts’ shortage caused by inaccurate prediction is solved.The specific work of this thesis is as follows:(1)In this thesis,the demand of parts in small batch and multi-batch production is studied.Combining its characteristics with the idea of centralized cutting,a method of predicting the parts demand of subsequent batches by Long Short-Term Memory network is proposed.(2)The demand prediction of subsequent batches is integrated,and the column generation method and knapsack problem algorithm are used to solve the multi-batch one-dimensional and two-dimensional guillotine cutting stock problems,and the layout is given.(3)The material shortage caused by prediction error is compensated.By classifying the batches that need to be compensated for cutting,the shortage of parts under the current batch is compensated,and the parts of subsequent batches are rolled to the next cutting task to meet the actual needs of parts.(4)The proposed LSTM-CSP model is constructed,and parts demand prediction performance experiment and cutting performance experiment verify its effectiveness and feasibility.The results show that the proposed method is better than the batch cutting method and inventory-based batch cutting method in one-dimensional optimization or two-dimensional guillotine cutting stock problem under the premise of high demand predicting accuracy.(5)A prototype system is developed for the proposed model.