A data-driven energy-saving planning method for machining processes

Abstract

The present invention discloses a data-driven method for energy-saving planning of machining processes, comprising: establishing a machining element energy consumption prediction model based on historical data collected from machine tools in a digital discrete workshop and numerical control programs, and solving the problem of energy-saving planning of machining processes in a digital discrete workshop in combination with the energy consumption prediction model. A mapping relationship is established between machine tool machining data, part feature machining elements, numerical control program parameters and machining energy consumption, a data set is obtained, and a designed neural network is trained to obtain an energy consumption prediction model of part feature machining elements. A machining process energy-saving planning problem model is established by optimizing the prediction model using the real-time machining data of workshop production, wherein the energy consumption of the machining elements is predicted by the trained prediction model; and the process planning model is optimized and solved by the designed intelligent optimization algorithm. The present invention can be used for process-assisted design of digital discrete workshops, improve workshop production efficiency, reduce energy consumption, and promote green and energy-saving production.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing and intelligent process planning, and in particular relates to a data-driven processing technology energy-saving planning method. Background Art

[0002] The job shop process planning problem is a type of sorting optimization problem that meets the requirements of resource allocation and process constraints. It is a typical NP-hard problem. In addition, process planning is an important link between part design and manufacturing. The quality of process planning results greatly affects the efficiency and cost of manufacturing. However, during the process planning process, it is difficult to accurately optimize the energy consumption of the entire process, resulting in an inability to determine the practicality of the process planning results in the actual processing environment. During the process planning process, a BP neural network is used to predict the processing energy consumption based on the historical data of the machine tool. The energy consumption target of the process planning is linked to the actual processing to obtain the optimal process route that meets the working conditions. Therefore, combining actual production equipment data, studying the energy-saving planning method and optimization method of this type of processing technology has important theoretical significance and engineering application value.

[0003] Neural networks are computing systems developed based on the simulation of human brain neural tissue. They possess essential characteristics of biological neural systems and, to a certain extent, reflect some aspects of human brain function, representing a simulation of biological systems. Neural networks are mathematical models that mimic the behavioral characteristics of animal neural networks and perform distributed parallel information processing. The BP neural network is a multi-layer feedforward neural network trained using the back propagation error algorithm. It boasts advantages such as large-scale parallelism, distributed processing, self-organization, and self-learning, and is widely used in fields such as prediction and reasoning. However, it also has limitations, including slow convergence and a tendency to fall into local minima.

[0004] Genetic algorithms are computational models of biological evolution that simulate the natural selection and genetic mechanisms of Darwinian evolution. They are a method for searching for optimal solutions by simulating natural evolutionary processes. They possess implicit parallelism and the ability to search across a global solution space, making them widely used in process planning. Classic genetic algorithms also offer excellent global optimization capabilities.

[0005] To control energy consumption and improve manufacturing efficiency during parts manufacturing, a data-driven energy-saving planning method for machining processes is proposed. This method enables intelligent machining process energy-saving planning that aligns with actual workshop conditions, enhancing the practical applicability of this theoretical research. Summary of the Invention

[0006] In response to the close connection between process planning and workshop production in digital discrete workshops, the present invention proposes a data-driven energy-saving planning method for machining processes, thereby realizing energy-saving planning for machining processes. Combining the numerical control program and the collected machine tool processing data, the parameters related to the machining energy consumption of the machining elements are mined, and an energy consumption prediction model for the machining elements is established to fit the energy consumption required for actual machining. A machining process energy-saving planning model is established based on the prediction model. Based on the model, a genetic algorithm is used for solving the problem. The problem of the existing process planning results being out of touch with the actual workshop production and the poor usability of the optimization results is solved, and intelligent process planning is realized to improve manufacturing efficiency and reduce energy consumption. To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A data-driven energy-saving planning method for machining processes includes the following steps:

[0008] Step 1: Based on the part processing element decomposition rules, decompose the processing methods of all feature matching of the part into independent processing units to obtain the processing element sequence for process sequencing optimization;

[0009] Step 2: Based on the real-time status data of the equipment operation, select appropriate processing resources for each processing element; perform sorting optimization with the minimum processing energy consumption as the optimization goal, and use the neural network prediction model to fit the energy consumption required for the actual processing of the parts to obtain the optimal energy-saving process route that is more in line with the actual working conditions of the workshop.

[0010] The implementation of step 2 is divided into three steps:

[0011] Step 1: Mining the processing data corresponding to the processing elements in the NC program, using the collected machine tool processing history data to establish a mapping model between the part feature processing elements, machine tool processing data, NC program and energy consumption;

[0012] The method to build the mapping model is as follows:

[0013] a) Extract the program segments of the machining process in the NC program;

[0014] b) Combine process information to minimize program segmentation of machining units, and process the machining data of the machine tool in the same way;

[0015] c) Use the machining unit as a link to connect the corresponding machine tool processing data and CNC program.

[0016] d) Complete the establishment of the mapping model.

[0017] Step 2: Analyze processing data based on the mapping model described in step 1, normalize the pre-processed data, train the neural network, establish an energy consumption prediction model for feature processing elements, and optimize the energy consumption prediction model using the collected real-time processing data;

[0018] The main process of building a prediction model is as follows:

[0019] a) Preprocessing the processing data collected by the machine tool and forming a sample set through normalization;

[0020] b) Establish an initial BP neural network and use the processed data for training;

[0021] c) Input the process case data into the BP neural network for testing to determine whether the error meets the standard.

[0022] d) Complete the adaptive energy consumption prediction model.

[0023] e) Use the real-time collected processing data to test the accuracy of the prediction model.

[0024] BP neural network design:

[0025] a) Analyze the data set used and determine the number of input and output nodes;

[0026] b) Determine the number of hidden layer nodes according to the empirical formula;

[0027] c) Set the Sigmoid function as the hidden layer transfer function and the linear function as the output layer function to prevent the output value from being limited between (0, 1) and (-1, 1);

[0028] d) The BP neural network training method is selected as the error back propagation algorithm, and a serial training method is used to avoid falling into the local optimum;

[0029] e) determining the initial weights in the neural network;

[0030] f) Normalize the training sample data and input it into the network for learning to obtain the required neural network.

[0031] Step 3: Optimize the process sorting with energy consumption as the optimization target, and predict the processing energy consumption required for each processing element through neural network; use intelligent optimization algorithm to optimize the process sorting with minimum energy consumption, and obtain the optimal process route with minimum energy consumption that meets the actual working conditions.

[0032] Design of genetic algorithm combined with BP neural network:

[0033] Genetic algorithms are used to optimize process planning. The processing resource constraints and the process constraints of the part are used as constraint information to be input into the algorithm optimization. The process steps are sorted and the optimal process planning result that meets all constraints is solved.

[0034] a) Coding: The chromosome coding method adopts integer coding based on processing elements. Each chromosome can represent the processing order of all processing elements of the workpiece.

[0035] b) Generation of initial solution: Most of the initial population individuals are generated by random generation method, and a small part of the initial population individuals are generated by heuristic rules.

[0036] c) Calculation of fitness function value: The target energy consumption E of the process plan corresponding to each individual is used as the fitness value of the population individual.

[0037] d) Genetic manipulation:

[0038] ① Selection: Based on the fitness value of each individual, the tournament selection method is used to select individual genes from the parent group and pass them on to the next generation;

[0039] ②Crossover: Perform single-point crossover with a certain crossover probability.

[0040] ③ Mutation: Two-point gene exchange mutation and insertion mutation are performed randomly with a certain mutation probability.

[0041] The beneficial effects of this invention include: utilizing machine tool processing big data to establish a process energy-saving planning model for digital discrete production workshops, with energy consumption as the optimization objective. Based on this model, a genetic algorithm combined with a BP neural network is proposed, which can effectively solve optimal or suboptimal solutions to process planning problems. This algorithm can be used for workshop part process design, improving workshop production efficiency, reducing energy consumption, and strengthening the connection between design and workshop manufacturing, thereby enhancing the practicality of process planning results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Flowchart of this method.

[0043] Figure 2 Develop a methodology for energy consumption mapping models.

[0044] Figure 3 This is the training flow chart of BP neural network. DETAILED DESCRIPTION

[0045] The technical solution of the present invention is described in detail below with reference to the accompanying drawings:

[0046] With the introduction and development of the concept of digital workshops, the data acquisition technology for CNC machine tools is becoming more and more sophisticated, and the collected processing data of CNC machine tools is becoming more and more accurate. There is a close connection between machine tool processing data and part characteristic parameters, and some of these data have a great impact on the energy consumption of process planning results, such as: machine tool spindle speed, cutting depth, feed rate and cutting speed. Factors that need to be considered in general discrete workshop process planning include: conflicts between the use of all processing resources, the rationality of the process sequence, process constraints, and whether the energy consumption required for the process results meets manufacturing requirements. Once the processing resources of the characteristic processing element are determined, the energy consumption required will also be determined accordingly, and the energy consumption predicted by data drive is.

[0047] The discrete workshop process planning problem can be described as: a single process part to be designed, N processing machines with different functions, M features of the part and related information, and the process constraints of the part. The process constraints depend on the relationship between the part features and the workshop processing conditions, and the impact of processing resources on the constraints must also be considered.

[0048] This invention analyzes historical data from CNC machine tools in a digital discrete workshop production process and provides a data foundation for accurately predicting the energy consumption required for characteristic processing elements. This energy consumption prediction model is obtained by training a BP neural network, and ultimately, a precise energy-saving planning model for the machining process is established using a genetic algorithm. The invention is further described below with reference to the accompanying figures and specific implementation methods of the method:

[0049] Step 1: Establish an energy consumption mapping model.

[0050] 1) Decomposition of NC program: The NC program segment represents the entire process of machining. The operation codes for different features are distinguished and the corresponding data information is extracted;

[0051] 2) Associated processing units: Divide the process into processing units based on the process steps and correspond them to the CNC programs;

[0052] 3) Processing machine tool data: dividing the collected machine tool data into operating parameters and processing results;

[0053] 4) Model establishment: Based on the machining unit and taking energy consumption as the result, the machine tool data, NC program and process information are connected to establish a mapping model;

[0054] Step 2: Establish a processing element energy consumption prediction model.

[0055] The BP neural network energy consumption prediction model is designed as follows:

[0056] 1) Set the number of network layers: BP neural network can contain one to multiple hidden layers. For the data used in this method, a single hidden layer can meet the requirements;

[0057] 2) Set the number of output and hidden nodes: The number of output layer nodes depends on the dimension of the input vector, the number of hidden layer nodes is determined by an empirical formula, and the number of output layer nodes is obtained based on the abstract model of the research problem;

[0058] 3) Transfer function selection: The hidden layer uses the Sigmoid function. The output layer uses a linear function to avoid limiting the output value due to the Sigmoid function.

[0059] 4) Selection of training method: Use the standard steepest descent method and serial training;

[0060] 5) Initial weights: BP neural network uses an iterative update method to determine weights, so it is necessary to determine an initial value, which is usually set to (-2.4 / F, 2.4 / F) or Between, F is the number of input nodes;

[0061] 6) Training model: Normalize the training data and input it into the network for learning. If the network converges successfully, the required BP neural network prediction model can be obtained;

[0062] Step 2: Establish a discrete workshop process planning model with energy consumption as the optimization target.

[0063] Now make the following assumptions:

[0064] 1) The energy consumption of each processing element is predicted in real time by the BP neural network and will not be changed by external factors once determined;

[0065] 2) Once the workpiece starts processing, it cannot be interrupted;

[0066] 3) All the processing elements of the workpiece can be processed on at least one machine tool;

[0067] 4) A machine tool can only process one processing element at a time;

[0068] 5) There is a priority relationship between all processing elements;

[0069] 6) The conversion energy consumption between different machine tools is only related to the distance;

[0070] 7) The energy consumption for adjusting tools on different machine tools is relatively low and is considered a fixed value;

[0071] The process constraints that the model must satisfy are:

[0072] 1) Typical process constraints

[0073] The typical process constraints are as follows: roughing first, then finishing; datum first, then others; surface first, then hole; main surface first, then secondary surface.

[0074] 2) Clustering constraint relationship

[0075] The principle of process concentration states that machining of each part surface should be concentrated in a small number of processes, with each process processing as much content as possible. Based on this principle, features of the same type, elements at the same machining stage for different features, and elements using the same machining resources should be processed together.

[0076] 3) Custom constraint relationships

[0077] Process personnel can set corresponding constraints based on the actual production environment and requirements of the workshop and specify the processing sequence between specific processing elements.

[0078] The relevant symbol definitions are shown in Table 1:

[0079] Table 1 Symbol definitions

[0080]

[0081] The processing constraints and calculation formula of this model are as follows:

[0082] Need to meet l i ≤l m and d i ≤d m (1)

[0083] 0≤b i,j ≤e i,j (2)

[0084] e i,j-1 ≤b i,j (3)

[0085]

[0086] t i,j =e i,j -b i,j (5)

[0087] Among them, formula (1) means that the available equipment must meet the workpiece processing space constraints, formula (2) means that the start processing time of the j-th processing element of the workpiece must be less than the completion time of its corresponding processing element, and the start processing time of the workpiece processing element is greater than or equal to 0, formula (3) means that the workpiece can only enter the next processing element for processing after the previous processing element is completed, formula (4) means that the workpiece can only be arranged to be processed on one device at each processing element, and formula (5) is the calculation formula for the processing time of the j-th processing element of workpiece i.

[0088] Step 2.1 Establish discrete workshop energy consumption model

[0089] In actual production, manufacturing energy consumption mainly includes three parts: processing energy consumption, conversion energy consumption and tool adjustment energy consumption. The processing energy consumption is predicted by the BP neural network. The other two parts of energy consumption are equal to the product of power and time:

[0090]

[0091] E y =P w ×t w (7)

[0092] E q =P e ×t e (8)

[0093] Step 2.2: For the above process planning model, a genetic algorithm is used to perform sorting optimization. The sorting problem in this model can be viewed as a path optimization problem, and the genetic algorithm is very capable in path optimization. Therefore, the present invention uses a genetic algorithm to optimize the sorting of processing elements. The main contents are as follows:

[0094] 1) Encoding: Data in the feasible space is used as the representation of the genetic algorithm population. The problem is abstracted into data, and the data in the feasible domain is then represented as the structured data of the genetic algorithm's gene string. This method uses a three-layer encoding approach: the first layer represents the process step, the second layer represents the machine tool used, and the third layer represents the tool used.

[0095] 2) Parameter initialization: Initialize parameters, including the number of iterations, population size, crossover probability, individual length, selection probability, and mutation probability;

[0096] 3) Randomly initialize the population and add the feature constraint analysis result matrix of the parts to reduce the generation of infeasible solutions;

[0097] 4) Fitness value calculation: The energy consumption corresponding to each individual is used as the fitness value of the population individual. The calculation formula is: Among them, E is predicted using the energy consumption prediction model;

[0098] 5) Selection operation: Based on the fitness value of each individual, the tournament selection method is used to select individual genes from the parent population to be inherited to the next generation;

[0099] 6) Crossover operation: A single-point crossover is performed with a certain crossover probability. First, two individuals are randomly selected from the population, and a point is randomly selected as the crossover position. The two chromosomes undergo a single-point crossover. After the crossover, the chromosomes may have redundant or missing processes. Therefore, the redundant processes of the workpiece are converted to the missing processes of the workpiece.

[0100] 7) Mutation Operation: Mutations are performed using random two-point swap mutations and insertion mutations. Two-point swap mutations randomly select two different gene positions in the parent chromosome and swap their gene values. Insertion mutations randomly generate two different gene positions, inserting the latter gene into the former gene's position, and then shifting the remaining genes backwards.

[0101] 8) If the termination condition is met, output the optimal solution; otherwise, return to step 4.

Claims

1. A data-driven processing technology energy-saving planning method, characterized in that: The following steps are involved: Step 1: Based on the part processing element decomposition rules, decompose the processing methods of all feature matching of the part into independent processing units to obtain the processing element sequence for process sequencing optimization; Step 2: Based on the real-time status data of the equipment operation, the appropriate processing resources are selected for each processing element. The sorting optimization is carried out with the minimum processing energy consumption as the optimization goal. The energy consumption required for the actual processing of the parts is fitted through the neural network prediction model to obtain the optimal energy-saving process route that better meets the actual working conditions of the workshop. The implementation of step 2 is divided into three steps: Step 1: Mining the processing data corresponding to the processing elements in the NC program, using the collected machine tool processing history data to establish a mapping model between the part feature processing elements, machine tool processing data, NC program and energy consumption; The method to build the mapping model is as follows: a) Extract the program segments of the machining process in the NC program; b) Combine process information to minimize program segmentation of machining units, and process the machining data of the machine tool in the same way; c) Using the machining unit as a link, connect the corresponding machine tool processing data and NC program; d) Complete the establishment of the mapping model; Step 2: Analyze processing data based on the mapping model described in step 1, normalize the pre-processed data, train the neural network, establish an energy consumption prediction model for feature processing elements, and optimize the energy consumption prediction model using the collected real-time processing data; The main process of building a prediction model is as follows: a) Preprocessing the processing data collected by the machine tool and forming a sample set through normalization; b) Establish an initial BP neural network and use the processed data for training; c) Input the process case data into the BP neural network for testing to determine whether the error meets the standard; d) Complete the adaptive energy consumption prediction model; e) Use real-time collected processing data to test the accuracy of the prediction model; BP neural network design: a) Analyze the data set used and determine the number of input and output nodes; b) Determine the number of hidden layer nodes according to the empirical formula; c) Set the Sigmoid function as the hidden layer transfer function and the linear function as the output layer function to prevent the output value from being limited between (0, 1) and (-1, 1); d) The BP neural network training method is selected as the error back propagation algorithm, and a serial training method is used to avoid falling into the local optimum; e) determining the initial weights in the neural network; f) Normalizing the training sample data and inputting it into the network for learning to obtain the required neural network; Step 3: Optimize the process sequence with energy consumption as the optimization target, and use the neural network to predict the processing energy consumption required for each processing element. Use the intelligent optimization algorithm to optimize the process sequence with minimum energy consumption and obtain the optimal process route with minimum energy consumption that meets the actual working conditions. Design of genetic algorithm combined with BP neural network: Use genetic algorithms to optimize process planning, taking the processing resource constraints and the existing process constraints of the part as the constraint information required for algorithm optimization, sorting the processes, and solving the optimal process planning result that meets all constraints; a) Coding: The chromosome coding method adopts integer coding based on processing elements. Each chromosome can represent the processing order of all processing elements of the workpiece; b) Generation of initial solution: Most of the initial population individuals are generated by random generation, and a small part of the initial population individuals are generated by heuristic rules; c) Calculation of fitness function value: The target energy consumption E of the process plan corresponding to each individual is used as the fitness value of the population individual; d) Genetic manipulation: ① Selection: Based on the fitness value of each individual, the tournament selection method is used to select individual genes from the parent group and pass them on to the next generation; ② Crossover: Perform single-point crossover with a certain crossover probability; ③ Mutation: Two-point gene exchange mutation and insertion mutation are performed randomly with a certain mutation probability.

Images