| For discrete manufacturing companies that adopt a multi-variety and small-batch flexible production mode,an effective production scheduling can improve production efficiency and save production costs.Meanwhile,the flexible job-shop scheduling problem(FJSP)is also a typical non-deterministic polynomial(NP)hard combinatorial optimization problem.The difficulties of FJSP are the variable condition,the computational complexity and the solving difficulty,which could be solved difficultly and accurately in polynomial time by utilizing mathematical method.Furthermore,many types of manufacturing resource and the dynamical degrading performance of equipment put forward high requirements for ensuring reliability and robustness of production scheduling.Therefore,the FJSP has become a hot topic in both academia and industry.As an important manufacturing resource in flexible manufacturing workshop,the cutting tool is also a necessary equipment for the machine tool to process workpiece.The remaining useful life of the tool is a critical index to evaluate its cutting performance and implement tool changing strategy.Although many researches have been carried out on the remaining useful life prediction of tool and FJSP and some achievements have been made,the existing research only focuses on the above two fields separately and rarely integrates the two fields.These make the tool failure during the machining process,resulting in some problems,such as the lower machining quality of the workpiece,machine breakdown,machine failure and delayed delivery,which will seriously affects production efficiency and production costs.Therefore,it is of great practical significance to regard tool life as a necessary constraint for the FJSP.However,the monitoring data of tool wear process has low signal noise ratio and strong time-variability,and it is impossible to collect the complete data of tool wear process from the beginning of the tool to the failure when the machining environment changes and the new tool is just put into use.Furthermore,the integration of the remaining useful life of tool into the FJSP will make the scheduling problem more constraints,higher complexity,more complicated decision-making factors,and more difficult to be solved.Thus,the higher requeirements are put forward to improve the solution quality and search efficiency of the optimization algorithm.These greatly increase the difficulty of the tool remaining useful life prediction and the FJSP.Therefore,to solve the FJSP with tool life constraint,it is important to develop a tool remaining useful life prediction approach with the incomplete data,and propose some high-prediction and efficient algorithms to solve scheduling problem,achieving the best scheduling solution in a short time.This dissertation focuses on the remaining useful life prediction of tool and the model construction and optimization algorithm design of FJSP.The main contributions of this dissertation are summarized as follows:(1)Due to the large differences in data feature distribution and wear trend in different tool wear stages,it is difficult for a single global model to accurately describe the tool wear process.To solve this problem,a multi-stage approach of tool remaining useful life prediction integrated with tool condition classification is proposed.Firstly,the multidimensional sensor signal is converted into a symmetrized dot pattern(SDP)image.By analyzing the influence of different SDP image parameters on the image characteristics,an adaptive selection method of SDP image parameters is established.According to the optimal image parameters,the SDP image clustering templates of different tool wear condition are constructed,and a new tool condition classification method based on the improved SDP is proposed.Then,by analyzing the data distribution characteristics of different tool wear stages,a tool wear prediction approach based on multi-covariance Gaussian process regression and its parameter optimization method are proposed.The different prediction models are assigned according to the tool condition classification results.Finally,compared with other methods,the effectiveness of the proposed multistage prediction method is validated.(2)When the machining environment changes and the new tool is just put into use,there is only sensor data from the beginning of the tool operation or a certain moment to the current moment(also called incomplete data),which makes it difficult to establish a remaining useful life prediction of tool.To address this problem,a novel method for predicting tool remaining useful life under incomplete data is proposed.Firstly,the relationship between the tool wear factor and the feature failure threshold is established by analyzing the tool wear mechanism.Meanwhile,an adaptive construction approach of the time-feature window is developed.On this basis,in order to solve the problem of tool wear trend representation,a processing method for abnormal time-feature window with abnormal data is designed.Then,a deep bidirectional long short-term memory is constructed and the prediction model is trained by using the compressed time-feature window.Combined with the tool wear factor,a multi-step ahead rolling prediction approach is presented,which can predict the wear trend of tool.Finally,compared with other methods,the effectiveness of the proposed tool remaining useful life prediction method is validated.(3)As the existing research of FJSP ignores the tool life constraint,the tool is easy to be damaged during the machining process,which leads to some problems,such as the lower quality of the workpiece,machine breakdown and delayed delivery.In order to solve the above problems,a mathematical model for FJSP with tool life constraint is formulated to minimize the total production cost.Firstly,an evaluation method for the diversity of solution sets is established,and the solution repair method and self-restart method are proposed so as to improve the quality of the solution.Then,according to the priority of the optimization goal,a heuristic-based two-stage approach based on the optimization of tardiness time and machining cost is investigated,which can instructively reduce the solution space and greatly enhence the solving efficiency of optimization algorithm.Finally,compared with other existing approaches,the superiority of FJSP model with tool life constraint and the heuristic-based two-stage approach is validated.(4)The existing research of multi-objective FJSP ignores the tool life constraint,and the multi-objective decision-making stage only considers low-dimensional information,which leads to the problem of poor feasibility of the scheduling scheme.To deal with the above problem,a mathematical model for multi-objective FJSP with tool life constraint is formulated to optimize the completion time,tardiness time and machining cost simultaneously.Firstly,by exploring the critical path considering the machine and the tool,a neighborhood search method considering tool life constraint is developed,which improves the diversity of the solution set.To solve multi-objective FJSP problem,a memetic algorithm is designed,and a set of non-dominated solutions are obtained.Secondly,a comprehensive weight calculation method is presented based on the highdimensional information,such as the internal correlation of the optimization goal,the preference of decision-makers,subjective knowledge and actual data.On this basis,a decision model for the non-dominated solutions is constructed to obtain the optimal production scheduling plan.Finally,compared with other methods,the effectivenss of the proposed memetic algorithm and multi-objective decision-making method is validated. |
PDF Full Text Request