A full-process production scheduling method and system for high-performance non-standard hydraulic cylinders

Abstract

The present invention discloses a full-process production scheduling method and system for high-performance non-standard hydraulic cylinders, the method comprising the following steps: obtaining the characteristics of each part and processing equipment, constructing a machining time prediction model, and predicting the machining time of each part; classifying the assembly process of the high-performance non-standard hydraulic cylinder according to the information in the assembly process card of the hydraulic cylinder, and predicting the assembly time of various types of high-performance non-standard hydraulic cylinders according to the predicted machining time of each part; establishing a multi-objective optimization function including minimization of assembly time and minimization of processing cost, solving the multi-objective optimization function, and obtaining the optimal production scheduling plan for high-performance non-standard hydraulic cylinders. The method of the present invention significantly improves the scientificity, accuracy and efficiency of production scheduling through integrated prediction and optimization.

Description

Technical Field

[0001] The present invention relates to the technical field of hydraulic cylinder production scheduling, and in particular to a full-process production scheduling method and system for high-performance non-standard hydraulic cylinders. Background Art

[0002] Hydraulic technology plays a vital role in aerospace, steel and metallurgy, engineering machinery, building materials and construction industries due to its high power output, fast response and high reliability. These industries rely on the high performance of large production line equipment, and high-performance non-standard hydraulic cylinders are the core components of these equipment. Their quality and reliability are directly related to the stable operation of the production line and the efficiency of spare parts maintenance.

[0003] In order to meet the production needs of high-performance non-standard hydraulic cylinders, the industry has developed a variety of technologies and methods. These technologies cover all stages from process design to full-process production control. Although these technologies have achieved automation and optimization to a certain extent, they are mainly focused on single production stages such as machining, heat treatment and assembly, and lack comprehensive scheduling and optimization of the entire production process.

[0004] Compared with the international advanced level, there is still a significant gap in the process design and production operation technology of high-performance non-standard hydraulic cylinders in China, resulting in unstable product quality, long delivery cycle and weak supporting capacity. The existing technology mainly focuses on a single production stage and fails to form a full-process production scheduling, which limits the global optimization of production efficiency and resource allocation. In addition, the existing production model relies on discrete machine tools and manual semi-automatic assembly, which makes the process design too dependent on experience, lacks precise theoretical support, and is difficult to accurately calculate the working hour quota. At the same time, in order to meet the requirements of product performance, position conditions, use media and ambient temperature of high-performance hydraulic cylinders, it is necessary to carry out non-standard design of hydraulic cylinder appearance dimensions, cylinder body materials, internal sealing system structure, etc. The complexity of non-standard product design and the difficulty of process planning further aggravate the challenges of production resource allocation, resulting in low production efficiency. Summary of the invention

[0005] The present invention proposes a full-process production scheduling method and system for high-performance non-standard hydraulic cylinders, which solves the problem that the prior art lacks production scheduling planning for the full process of high-performance non-standard hydraulic cylinders, resulting in low production efficiency.

[0006] In order to solve the above technical problems, the present invention provides a full-process production scheduling method for high-performance non-standard hydraulic cylinders, comprising the following steps:

[0007] Step S1: Obtain the characteristics of each part and processing equipment, build a machining time prediction model, and predict the machining time of each part;

[0008] Step S2: classifying the assembly process of the high-performance non-standard hydraulic cylinder according to the information in the assembly process card of the hydraulic cylinder, and predicting the assembly time of various types of high-performance non-standard hydraulic cylinders according to the predicted machining hours of each part;

[0009] Step S3: Establish a multi-objective optimization function including minimization of assembly time and minimization of processing cost, solve the multi-objective optimization function, and obtain an optimal production scheduling plan for high-performance non-standard hydraulic cylinders.

[0010] Preferably, step S1 comprises the following steps:

[0011] Step S11: the characteristics of the equipment include cutting speed, feed rate and cutting depth, and the characteristics of the part include workpiece size and material hardness;

[0012] Step S12: training a neural network model based on historical data of equipment characteristics and part characteristics and historical data of machining hours;

[0013] Step S13: Obtain feature data of the parts to be processed and the available equipment, input the trained neural network model, and obtain the predicted machining time for each part.

[0014] Preferably, in step S1, the predicted machining time is optimized by the following expression:

[0015]

[0016] In the formula, H i is the machining time of optimized part i; Tmq i is the machining time of part i; λ is the learning index; S(j,i) is the similarity between part j and part i; β is the control parameter of truncated learning effect; n is the total number of similar features; d k is the weight of similar feature i; r k (i, j) is the similarity value between parts i and j under similar feature k.

[0017] Preferably, the step S2 of classifying the assembly process of the high-performance non-standard hydraulic cylinder according to the information in the assembly process card of the hydraulic cylinder comprises the following steps:

[0018] Step S201: obtaining text data in a plurality of assembly process cards as a training data set, performing text segmentation on the training data set, and removing stop words in the training data set after segmentation to obtain segmentation data;

[0019] Step S202: Calculate the feature value of each word in the word segmentation data, and select words whose feature value is greater than a set feature value threshold to form a feature word library;

[0020] Step S203: Calculate the frequency of each category of assembly process cards in the training data set as the prior probability of the category; for each category, calculate the conditional probability of each feature word appearing in the category;

[0021] Step S204: according to the vocabulary in the feature word library, the posterior probability that the assembly process card to be classified belongs to each category is calculated using the Bayesian formula, and the category with the highest posterior probability is selected as the predicted category of the assembly process card to be classified.

[0022] Preferably, step S2 adopts a proximal strategy optimization algorithm to predict the assembly time of various types of high-performance non-standard hydraulic cylinders according to the predicted machining hours of each part, including the following steps:

[0023] Step S211: Initializing the parameters of the proximal strategy optimization algorithm and the environmental parameters, wherein the environmental parameters include the number of workpieces, the number of machines, the process and processing time of each workpiece;

[0024] Step S212: obtaining an initial state, inputting the initial state into a policy network, outputting a probability distribution of selecting each workpiece, and selecting an action from the action space according to the probability distribution;

[0025] Step S213: Execute the selected action, update the state matrix and instant reward;

[0026] Step S214: storing the current state, action, reward and new state into the experience playback buffer;

[0027] Step S215: Calculate the discounted cumulative return for each time step using the stored rewards, and update the parameters of the policy network using the discounted cumulative return and the probability distribution of the policy network output through the policy gradient method; Update the parameters of the value network by minimizing the prediction error of the state;

[0028] Step S216: Repeat steps S212 to S215 until the set maximum number of iterations is reached.

[0029] Preferably, the expression for minimizing the assembly time in step S3 is:

[0030] min C max =max(E p ,p=1,2,...,|p|);

[0031] E p =S p +t aqp ;

[0032]

[0033] In the formula, C max is the assembly completion time of the last product; E p is the assembly completion time of product p, S p is the assembly start time of product p; t aqp is the predicted assembly time of product p; L p is the latest arrival time of all workpieces that make up product p; Z p,p' is a decision variable. If all parts of product p arrive at the factory, the previous product p′ has not been assembled, then it is 1; otherwise, it is 0. p′ is the assembly completion time of the immediate predecessor product p′ of product p; μ j,p is a decision variable, which is 1 if workpiece j belongs to product p, otherwise it is 0; θ j,h is a decision variable, which is 1 if product p is transported by vehicle h, otherwise it is 0; S h is the start transportation time of vehicle h; η j,f is a decision variable, which is 1 if product p is produced in factory f, otherwise it is 0; f is the transportation time from factory f to the assembly plant; E j is the processing completion time of workpiece j.

[0034] Preferably, the expression for minimizing the processing cost in step S3 is:

[0035] min TC=TPC+TICj+TAC+TICp+TDC+TTC;

[0036]

[0037] Where TPC is the cost of the processing process; F is the total number of factories; M is the total number of machines in factory f; n is the total number of workpieces that need to be processed by machine m; r j is the total number of processes for workpiece j; k j is the processing cost per unit time; x j,i,f,m is a decision variable, which is 1 if the i-th process of workpiece j is processed on the m-th machine in factory f, otherwise it is 0; t j,i,f,m is the time required for the mth machine in factory f to process the i-th process of workpiece j; TICj is the inventory cost of the workpiece; N is the total number of products; v j is the unit time inventory of workpieces; μ j,p is a decision variable, which is 1 if workpiece j belongs to product p, otherwise it is 0; Sa p is the assembly start time of product p; E j is the processing completion time of workpiece j; TAC is the cost of the assembly process; k p is the assembly cost per unit time; t pis the assembly time of product p; TICp is the inventory cost of the product; H is the total number of vehicles; v p is the unit time inventory cost of the product; σ p,h is a decision variable, which is 1 if product p is assigned to vehicle h, otherwise it is 0; Ea p is the assembly completion time; TDC is the transportation cost; ω is the fixed cost of the vehicle; U f,c A collection of factories and customers' locations; k h is the transportation cost per unit time; is a decision variable, which is 1 if d is a destination in the delivery process of vehicle h, otherwise it is 0; d,d' is a decision variable, which is 1 if location d' is delivered immediately after location d, otherwise it is 0; t d,d' is the transportation time from location d' to location d; TTC is the delay cost; C is the total number of customers; k T is the delay cost per unit time; p,c is a decision variable, which is 1 if product p belongs to customer c, otherwise it is 0; T c The time it takes for a vehicle to deliver the product to customer c; E c is the delivery deadline for customer c.

[0038] The present invention also provides a full-process production scheduling system for high-performance non-standard hydraulic cylinders, which is implemented based on the above-mentioned full-process production scheduling method for high-performance non-standard hydraulic cylinders, including a machining time prediction module, an assembly time prediction module and a production scheduling optimization module;

[0039] The machining time prediction module is used to train a machining time prediction model based on historical machining data, and input the machining data of the workpiece to be predicted into the trained machining time prediction model to obtain the machining time of the workpiece to be predicted;

[0040] The assembly time prediction module: trains an assembly time prediction model based on historical assembly data, classifies various assembly tasks of high-performance non-standard hydraulic cylinders, inputs the classified assembly data into the trained assembly time prediction model, and obtains the assembly time of each assembly task;

[0041] The production scheduling optimization module formulates a production scheduling plan based on the predicted time of machining and assembly, and dynamically adjusts the production scheduling plan based on real-time production data and external changes.

[0042] Preferably, the machining time prediction module further comprises a machining time prediction optimization module, wherein the machining time prediction optimization module optimizes the machining time prediction value of each part by calculating the similarity between the parts.

[0043] Preferably, the assembly time prediction module includes an assembly process classification module, which constructs a feature word library by extracting feature values ​​of text data in a plurality of assembly process cards, and classifies the assembly process cards using a Bayesian algorithm based on the feature word library.

[0044] The benefits of the present invention include at least:

[0045] 1. By building a machining time prediction model based on a large amount of feature data, the machining time of each part can be predicted more accurately, which is more scientific and accurate than the traditional method that relies on empirical estimation. By analyzing the information in the assembly process card to classify the assembly process, the process management is more systematic, which helps to improve the control ability of complex assembly processes. The prediction of machining and assembly time is integrated to form a unified prediction process, which helps to optimize the whole process production scheduling.

[0046] 2. By establishing a multi-objective optimization function including minimization of assembly time and minimization of processing cost, it is possible to consider multiple production scheduling objectives at the same time and achieve a more comprehensive and balanced scheduling solution. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A schematic diagram of a method flow of an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of a product MBD data set according to an embodiment of the present invention;

[0049] Figure 3 It is a framework diagram of the process MBD model of an embodiment of the present invention;

[0050] Figure 4 A schematic diagram of a machined PPR model according to an embodiment of the present invention;

[0051] Figure 5 A schematic diagram of the assembly process classification process of an embodiment of the present invention;

[0052] Figure 6 It is a structural schematic diagram of the machining flexible job shop scheduling model IPPS according to an embodiment of the present invention;

[0053] Figure 7 It is an example of the network graph OR node control function in the embodiment of the present invention;

[0054] Figure 8 It is a characteristic table and network diagram of the parts in the embodiment of the present invention;

[0055] Fig. 9 A parts process network diagram of a non-standard hydraulic cylinder according to an embodiment of the present invention;

[0056] Fig.10 Schematic diagram of a method for solving the joint scheduling optimization problem of distributed production assembly and delivery in an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following is a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0058] Through a comprehensive and in-depth study of the full-process production process and characteristics of non-standard hydraulic cylinders, a full-process production scheduling optimization framework for high-performance non-standard hydraulic cylinders is constructed. This framework targets the core characteristics of the full-process production of high-performance non-standard hydraulic cylinders and the scheduling optimization needs, and makes a global and scientific plan for the full-process production scheduling of high-performance non-standard hydraulic cylinders from the perspectives of mechanism, framework, method and application. The framework includes five levels of content: full-process production process, production mode and stage division, labor quota model, production scheduling model mapping and optimization solution, theoretical verification and practical application of hydraulic cylinder full-process production scheduling methods. These contents are both relatively independent and interrelated. This framework targets the core characteristics of the full-process production of high-performance non-standard hydraulic cylinders and the scheduling optimization needs, and makes a global and scientific plan for the full-process production scheduling of high-performance non-standard hydraulic cylinders from the perspectives of mechanism, framework, method and application.

[0059] For the full-process production scheduling optimization framework of high-performance non-standard hydraulic cylinders, the embodiment of the present invention provides the following Figure 1 The full-process production scheduling method of a high-performance non-standard hydraulic cylinder shown in the figure includes the following steps:

[0060] Step S1: Obtain the characteristics of each part and processing equipment, build a machining time prediction model, and predict the machining time of each part.

[0061] Specifically, based on the product digital model and the model-based definition method MBD, according to the existing non-standard hydraulic cylinder digital process design system and machining process card, the data elements that the machining process design and process manufacturing process rely on and are related to are comprehensively analyzed, and the following are extracted and established: Figure 2 The product MBD dataset shown.

[0062] Study the relationship between machining process and procedures, steps and geometric features, and establish Figure 3The machining process MBD model framework is shown in Figure 1. The process MBD model includes the product MBD model, the process MBD model and the process attributes, with the process MBD model as the core. The process of a blank becoming a finished product after cold working is essentially the process of the blank's three-dimensional geometric evolution, and also the process of the process MBD model evolution. A part needs to go through n processes from a blank to a finished part, and each process contains P i The processing feature G is cut in each process. ij Therefore, the evolution model of process MBD can be formally expressed as:

[0063] IPM 1 =M S ;IPM i =IPM i-1 -G ij (i-2,3,...,n-1);

[0064] Among them, IPM i represents the process model of process i; M s Represents the blank model; G ij is the jth processing feature of the i-th process, which is composed of geometric elements, related annotations and machining process attribute information.

[0065] The process MBD model is centered on processing features and can be expressed as:

[0066]

[0067] Among them, IPM bdi is the process model of process i with processing features as the core; Note it is the annotation information contained in the MDB model of the i-th process, such as heat treatment method, roughness grade, etc.; Mark ik It is the marking information contained in the MDB model of the i-th process, such as clamping reference, reference positioning, etc.; is the attribute information of the MDB model of the i-th process, such as process number, process name, process content, machine tool model, tool, etc.; F ijo The geometric elements and auxiliary geometric elements on the jth processing feature of the MBD model of the i-th process; N ijp All the annotation information of the jth processing feature of the MBD model of the i-th process; is the machining process attribute information of the jth machining feature of the MBD model of the i-th process, such as machining method, machine tool, tool and other information; m is the sum of geometric elements; n is the sum of machining processes; r is the sum of marking information contained in the MBD model of the process; s is the sum of machining process attributes of the machining features.

[0068] Based on the collected data sets and the constructed evolution process model, the product layer, process layer, and resource layer are divided to establish Figure 4 The three-layer resource, product and process PPR model of machining is shown in Figure 1. Based on this model, the machining time prediction model can be derived. The machining process quota time (Tmq) of all machining process time, equipment operation time, etc. can be expressed as Tmq = {Tmq 1 ,...,Tmq n By quoting the working hours, basic data can be provided for the production scheduling of the entire manufacturing process. The construction of the machining working hours prediction model specifically includes the following steps:

[0069] Step S11: Collect data related to machining process time including cutting parameters, workpiece characteristics, equipment information, etc. as input of the neural network. In the embodiment of the present invention, cutting speed, feed rate, cutting depth, workpiece size and material hardness are used.

[0070] Step S12: Normalize the input data to a specific range [0,1] to improve the model training effect:

[0071]

[0072] Among them, x normalized is the normalized data; x is the original data, x min 、x max are the minimum and maximum values ​​in the data respectively.

[0073] Step S13: Setting the parameters of the neural network. The neural network in the embodiment of the present invention adopts a multi-layer perceptron.

[0074] The number of neurons in the input layer of the multilayer perceptron is set according to the number of input features. The number of neurons in the input layer should correspond to the number of features of the input data. Each input neuron represents a specific input feature, so that the model can directly receive and process these feature information. Therefore, 5 neurons are set in the embodiment of the present invention.

[0075] Generally speaking, the more complex the problem and the larger the amount of data, the more hidden layers and neurons are needed. For high-precision, multi-process non-standard hydraulic cylinder processing, three hidden layers are required. In the embodiment of the present invention, the number of neurons in each hidden layer is set to 64, 32, and 16 respectively. When there are thousands or even tens of thousands of data, the number of neurons in each hidden layer can be adjusted between 32 and 128 according to the specific situation. Using a fully connected layer as a hidden layer, each neuron receives the output of all neurons in the previous layer, and linearly combines them through weights and biases, and then applies the activation function ReLU.

[0076] The output layer of the embodiment of the present invention is provided with only one neuron, which represents the predicted working hour quota.

[0077] Step S14: Collect historical data of equipment features and part features and historical data of machining hours, and divide them into training set, validation set and test set. The training set is used to train the model, the validation set is used to adjust hyperparameters and monitor the performance of the model, and the test set is used to finally evaluate the generalization ability of the model. The data is divided into 70% training set, 20% validation set, and 10% test set.

[0078] The mean square error is used as the loss function to optimize the parameters of the multilayer perceptron. The multilayer perceptron is evaluated on the test set, and the output loss value is the mean square error value. The mean square error is one of the commonly used evaluation indicators in regression problems. It reflects the degree of difference between the model's predicted value and the true value by calculating the average value of the square difference between the predicted value and the true value. The smaller the value, the better the prediction effect of the model.

[0079] The embodiment of the present invention uses the adaptive moment estimation optimization algorithm Adam to optimize the parameters of the multilayer perceptron until a predetermined number of training rounds is reached or the loss no longer decreases.

[0080] Step S15: Obtain the characteristic data of the parts to be processed and the available equipment, input the trained neural network model, and obtain the predicted machining time for each part.

[0081] By inputting information such as workpiece features and processing parameters of parts into the trained multi-layer perceptron, the corresponding machining time prediction value can be obtained.

[0082] After obtaining the predicted machining time of each part, according to the above-mentioned process planning and machining integrated scheduling problem, the influence of the similarity S(j,i) of all previous workpieces on the machining time of workpiece j is comprehensively considered. This cumulative method can better reflect the role of learning effect in the entire machining process, that is, as the number of workpieces to be processed increases, the machining experience of the previous workpieces will affect the machining time of the current workpiece j. The predicted machining time of each part is optimized by the following expression:

[0083]

[0084] In the formula, H i is the machining time of optimized part i; Tmq i is the machining time of part i; λ is the learning index, λ≤0; S(j,i) is the similarity between part j and part i; β is the control parameter of truncated learning effect; n is the total number of similar features; d k is the weight of similar feature i; r k(i, j) is the similarity value between parts i and j under similar feature k; r is the learning rate; β is the control parameter of truncated learning effect, 0<β<1.

[0085] Step S2: Analyze the assembly process card data, model and big data characteristics of the hydraulic cylinder, establish a non-standard hydraulic cylinder process rapid classification model based on process digital model mining, build an assembly man-hour prediction model based on advanced deep learning algorithms, run the prediction model, compare, feedback and correct the actual man-hour data, realize online accurate prediction of assembly man-hours, and provide man-hour data for full-process assembly scheduling optimization. Figure 5 As shown, the classification of the assembly process of high-performance non-standard hydraulic cylinders according to the information in the assembly process card of the hydraulic cylinder includes the following steps:

[0086] Step S201: sort out the types of hydraulic cylinders and their assembly process, obtain text data in multiple assembly process cards and assembly digital models as training data sets, perform text segmentation on the training data sets, and remove stop words in the training data sets after segmentation to obtain segmentation data.

[0087] Step S202: Use the term frequency-inverse document frequency (TF-IDF) algorithm to calculate the feature value of each word in the word segmentation data, select words with feature values ​​greater than the set feature value threshold to form a feature word library, and calculate the feature value of each word in the word segmentation data using the following formula:

[0088]

[0089] In the formula, f ij is the TF-IDF value of word i in process j; ω ij is the word frequency TF of word i in process j; is the inverse document frequency IDF of word i in the corpus; n ij is the number of times word i appears in process j; k j is the number of words in process j; N is the total number of processes in the corpus; N i is the number of processes in the corpus that contain word i.

[0090] Where N i+1 This is to prevent the denominator from being zero, that is, all processes in the corpus do not contain word i; f ij The value represents the importance of word i in process j, that is, f ij The larger the value, the more important word i is in process j; The value represents the uncommonness of word i in the corpus, that is, The larger the value, the less common word i is in the corpus.

[0091] Step S203: Calculate the frequency of each category of assembly process cards in the training data set as the prior probability of the category; for each category, calculate the conditional probability of each feature word appearing in the category.

[0092] Step S204: According to the vocabulary in the feature word library, the posterior probability that the assembly process card to be classified belongs to each category is calculated using the Bayesian formula, and the category with the highest posterior probability is selected as the predicted category of the assembly process card to be classified.

[0093] Build a non-standard hydraulic cylinder assembly man-hour prediction model based on deep learning and other algorithms: ① Use curve fitting to predict man-hours, use Matlab to draw smooth curves and extract fitting functions; ② Use general deep learning to predict man-hours; ③ Convolutional neural network man-hour prediction, use assembly man-hours together with hydraulic cylinder categories, process information, current scheduling plan information, and historical and real-time status information of workshop equipment as input data for model regression solution. Compare the gap between the man-hours predicted by these three methods and the actual man-hours, and finally use the optimal convolutional neural network and other solutions. The assembly man-hour prediction model based on deep reinforcement learning built in the embodiment of the present invention includes an action network and a value network:

[0094] (1)Actor

[0095] The task of the action network is to generate a selection probability for each workpiece, thereby determining the next step selection. Specifically, the state matrix S is input into the convolutional neural network, and features are first extracted through two convolutional layers. Each convolutional layer uses a 3×3 convolution kernel, and the step size and padding are both 1. This can capture the dependency between adjacent elements without changing the size of the feature map. Each convolutional layer uses the ReLU activation function to enhance nonlinear capabilities.

[0096] Next, after the convolutional layer processing, the output feature map is flattened and input into two fully connected layers for further processing, outputting the raw score of each action. Finally, the Softmax function is used to convert the output raw score into the selection probability distribution π for each artifact θ (a|s). These probability values ​​will be used to guide the behavior of the agent and help it make the best decision in the environment. This network design can capture the complex relationships in the state matrix, dynamically adjust the priority of artifacts, and improve the accuracy and flexibility of decision-making. During the training process, through interaction with the environment and updating of policy gradients, the network will continuously optimize its parameters to improve the overall performance of the agent.

[0097] (2) Value Network (Critic)

[0098] The value network is used to evaluate the value of the current state and provide feedback to the action network. Its structure is similar to that of the action network, but the final output is a scalar value, which represents the value of the state V(S). Through two convolutional layers and two fully connected layers, the value network can effectively extract state features and accurately evaluate the pros and cons of different states. This evaluation provides reliable feedback to the action network and guides its optimization selection strategy. Specifically, the output of the value network, that is, the state value, is used to calculate the gradient of the probability distribution of the action network output action, thereby updating the parameters of the action network. This feedback mechanism based on the value network enables the action network to better adapt to the environment and improve the accuracy of decision-making.

[0099] In the embodiment of the present invention, a proximal strategy optimization algorithm is used to predict the assembly time of various types of high-performance non-standard hydraulic cylinders according to the predicted machining hours of each part, including the following steps:

[0100] Step S211: Initialize the policy network and value network of the proximal policy optimization algorithm, and set the optimizer and hyperparameters, including the learning rate α, discount factor γ, etc., define the action space A and the state space S, and initialize the environmental parameters, including the number of workpieces, the number of machines, the process of each workpiece, and the processing time.

[0101] Step S212: Call the reset() method of the environment to reset the state space and obtain the initial state S 0 ;

[0102] Step S213: Set the current state S t Input policy network, output probability distribution of each workpiece π(a|s t ), select action a from the action space according to the probability distribution output by the policy network t , that is, select a workpiece j for the next operation.

[0103] Step S214: Submit the selected workpiece j and the corresponding operation to the environment for execution. The environment determines the best machine and operation time based on the current state and the selected action, and updates the state matrix S. The environment returns the new state S after execution. t+1 、Instant Rewards t And a flag indicating whether the current process has ended.

[0104] Step S215: Store the current state, action, reward and new state into the experience replay buffer.

[0105] Step S216: Calculate the discounted cumulative return R for each time step using the stored rewards tThrough the policy gradient method, the discounted cumulative return and the probability distribution of the policy network output are used to update the policy network parameters θ, so that actions that can obtain higher cumulative returns are given priority; by minimizing the prediction error of the state, the value network parameters φ are updated so that it can accurately evaluate the value of each state.

[0106] Step S217: Repeat steps S212 to S216 until the set maximum number of iterations is reached.

[0107] After a period of training, the performance of the algorithm can be evaluated: by running the current strategy in the test environment, recording the completion time of each workpiece and the fundamental indicators of production scheduling for the overall scheduling, the performance of the strategy can be evaluated. Based on the evaluation results, the hyperparameters of the algorithm are adjusted to further optimize the algorithm performance. Based on the adjusted parameters and model, the process of environmental interaction and model update is continued until the expected scheduling performance is achieved.

[0108] The algorithm process gradually optimizes the scheduling strategy through four stages: initialization, environment interaction, model update, and evaluation adjustment, so that the reinforcement learning algorithm can effectively learn and improve the scheduling efficiency of the job shop. This method not only fully utilizes the powerful capabilities of deep reinforcement learning, but also ensures that the algorithm performs well in complex scheduling problems through detailed environment interaction and feedback mechanisms.

[0109] Finally, the predicted result of assembly time is Taq = {Taq 1 ,Taq 2 ,...,Taq n}, embed it into the manufacturing execution system (MES), input data such as the number of workpieces, the number of machines, the process and processing time of each workpiece for online prediction, and use it for real-time adjustment of production scheduling.

[0110] Step S3: Establish a multi-objective optimization function including minimization of assembly time and minimization of processing cost, solve the multi-objective optimization function, and obtain the optimal production scheduling plan for high-performance non-standard hydraulic cylinders.

[0111] Specifically, in order to solve the planning and scheduling problems of flexible processes, a hierarchical modeling method is first used to decompose the integrated process planning and scheduling (IPPS) problem into two sub-problems: process planning problem and job shop scheduling problem (JSP). On this basis, in order to further study the multi-objective process planning and machining integrated scheduling optimization model, a corresponding IPPS scheduling model can be constructed for different production scenarios. In addition to taking the completion time as the basic optimization goal, the embodiment of the present invention also constructs a multi-objective system, including delay penalties, machine load, production cost, workpiece quality, and resource utilization, etc., to achieve more comprehensive and detailed production process optimization.

[0112] In order to cope with the dynamic changes that may occur in the production process, the embodiment of the present invention comprehensively analyzes the types of dynamic events related to workpieces, equipment, time, and processes. A dynamic scheduling driving strategy is designed for these dynamic events that may disturb production. Events are divided into different levels according to the degree of disturbance, and the mapping relationship between disturbance events and rescheduling strategies and their algorithms is explored, thereby establishing a variety of dynamic scheduling strategies, such as insertion scheduling, event-driven scheduling, periodic event hybrid driven scheduling, reorganization scheduling, and cross scheduling. The following steps are included:

[0113] Step S31: To perform hierarchical modeling on the flexible process planning and scheduling problem, the IPPS problem needs to be first decomposed into the process planning problem and the job shop scheduling problem JSP.

[0114] IPPS hierarchical modeling: Based on the idea of ​​flexibility decomposition, by decomposing the overall flexibility in the problem into two layers, a Figure 6 The hierarchical optimization IPPS model shown in Figure 1. The first layer of the model mainly determines the process selection subproblem and the machine selection subproblem.

[0115] Compared with the traditional process planning problem, the first layer of the model aims to minimize the total processing time of each workpiece. In the embodiment of the present invention, the process selection, i.e., the branch selection and machine selection of the OR node, is optimized by introducing the decision variable parameter o in the characterization method of high-performance non-standard hydraulic cylinder process planning based on the process network diagram. ijk To control the function of the OR node on the process node, use the variable z jk To describe the selection state of the OR node branch, the variable x i To describe the selection status of the process. Figure 7 As shown in Figure 2, if process i is controlled by the kth branch of the jth OR node, o ijk =1; otherwise, o ijk = 0. If the kth branch of the jth OR node is selected, then z jk =1, otherwise z jk =0; if process i is selected, then x i =1, otherwise x i = 0. The control logic of the OR node is summarized as follows: if and only if all branches controlling process i are selected, process i will be selected; otherwise, as long as one of the control branches is not selected, process i will not be selected. Based on the above operation, we can get Figure 8 The feature table and process network diagram of a part are shown in Figure 1. By summarizing the process network diagrams of all parts, we can get the following: Fig. 9The process network diagram of non-standard hydraulic cylinder parts shown in the figure inputs the process information and processing machine information contained in this semi-finished process route into the second layer of the model. The second layer of the model determines the order between all processes, with the goal of minimizing the maximum completion time, and obtains the scheduling solution for the overall IPPS problem.

[0116] Step S32: Study the multi-objective process planning and machining integrated scheduling optimization model. On the basis of hierarchical modeling, build the IPPS scheduling model under multiple scenarios. According to the predicted assembly time obtained above, establish the basic optimization goal of the fundamental production scheduling indicator makespan, and then establish a multi-objective system such as delay penalty, machine load, production cost, workpiece quality, and resource utilization.

[0117] In the embodiment of the present invention, a multi-objective optimization problem including minimization of assembly time and minimization of processing cost is constructed, wherein the expression for minimizing assembly time is:

[0118] min C max =max(E p ,p=1,2,...,|p|);

[0119] E p =S p +t aqp ;

[0120]

[0121] In the formula, C max is the assembly completion time of the last product; E p is the assembly completion time of product p, S p is the assembly start time of product p; L p is the latest arrival time of all workpieces that make up product p; S h is the start transportation time of vehicle h; θ j,h is a decision variable, which is 1 if product p is transported by vehicle h, otherwise it is 0; E j is the processing completion time of workpiece j; E p′ is the assembly completion time of the immediate predecessor product p′ of product p; Z p,p' is a decision variable. If all parts of product p arrive at the factory, the previous product p′ has not been assembled, then it is 1; otherwise, it is 0. aqp is the predicted assembly time of product p; μ j,p is a decision variable, which is 1 if workpiece j belongs to product p, otherwise it is 0; η j,f is a decision variable, which is 1 if product p is produced in factory f, otherwise it is 0; f is the transportation time from factory f to the assembly plant.

[0122] The expression for minimizing processing cost is:

[0123] min TC=TPC+TICj+TAC+TICp+TDC+TTC;

[0124]

[0125]

[0126] Where TPC is the cost of the processing process; F is the total number of factories; M is the total number of machines in factory f; n is the total number of workpieces that need to be processed by machine m; r j is the total number of processes for workpiece j; k j is the processing cost per unit time; x j,i,f,m is a decision variable, which is 1 if the i-th process of workpiece j is processed on the m-th machine in factory f, otherwise it is 0; t j,i,f,m is the time required for the mth machine in factory f to process the i-th process of workpiece j; TICj is the inventory cost of the workpiece; N is the total number of products; v j is the unit time inventory of workpieces; μ j,p is a decision variable, which is 1 if workpiece j belongs to product p, otherwise it is 0; Sa p is the assembly start time of product p; E j is the processing completion time of workpiece j; TAC is the cost of the assembly process; k p is the assembly cost per unit time; t p is the assembly time of product p; TICp is the inventory cost of the product; v p is the unit time inventory cost of the product; σ p,h is a decision variable, which is 1 if product p is assigned to vehicle h, otherwise it is 0; Ea p is the assembly completion time; TDC is the transportation cost; ω is the fixed cost of the vehicle; H is the total number of vehicles; U f,c A collection of factories and customers' locations; h is the transportation cost per unit time; is a decision variable, which is 1 if d is a destination in the delivery process of vehicle h, otherwise it is 0; d,d' is a decision variable, which is 1 if location d' is delivered immediately after location d, otherwise it is 0; t d,d' is the transportation time from location d' to location d; TTC is the delay cost; k T is the delay cost per unit time; p,c is a decision variable, which is 1 if product p belongs to customer c, otherwise it is 0; C is the total number of customers; T c The time it takes for a vehicle to deliver the product to customer c; Ec is the delivery deadline for customer c.

[0127] Step S33: Comprehensively analyze the types of dynamic events related to workpieces, equipment, time, and processes. For dynamic events that disturb production, study dynamic scheduling driving strategies, divide disturbance events into different levels based on the degree of disturbance, explore the mapping relationship between disturbance events and rescheduling strategies and their algorithms, and establish dynamic scheduling strategies such as insertion scheduling, event-driven scheduling, periodic event hybrid driven scheduling, reorganization scheduling, and cross scheduling.

[0128] Step S34: Optimizing and solving the single-objective or multi-objective, static or dynamic IPPS scheduling model in order to find the best production scheduling solution.

[0129] Specifically, for the multi-objective optimization allocation problem in the optimization model, the embodiment of the present invention studies the comprehensive multi-objective optimization configuration method from multiple angles, and comprehensively adopts multi-objective weight evaluation based on FAHP, target solution based on finding Pareto optimality, dimensionality reduction based on PCA dominant mechanism, and Nash equilibrium game strategy to perform multi-objective optimization allocation and conflict resolution, and compare and evaluate them to obtain the best multi-objective optimization allocation.

[0130] like Fig.10 As shown, in order to obtain the overall solution of the intelligent optimization algorithm for the joint scheduling optimization model of distributed hydraulic cylinder production, assembly and distribution, the embodiment of the present invention studies the algorithm improvement mechanism from five aspects: algorithm self-improvement, algorithm hybrid optimization, hybrid local optimization, algorithm parameter optimization, and algorithm efficiency optimization, so as to select hybrid algorithms and optimization strategies with high efficiency, stability and strong optimization capabilities; in terms of multi-objective optimization allocation, a dominant mechanism based on PCA is adopted; in terms of performance evaluation and improvement of hybrid optimization algorithms, single comparative tests, hybrid comparative tests, approximate optimal solution comparisons, solution time comparisons, and solution stability comparisons are carried out to verify the performance of the proposed algorithm and the effectiveness of the model.

[0131] Finally, based on the existing discrete manufacturing PLM / ERP / MES integrated platform and the full-process business data of hydraulic cylinder process production and assembly, combined with simulation optimization tools and development tools, we can develop a full-process production scheduling optimization algorithm library for hydraulic cylinders, and conduct platform development and method theory verification based on the full-process production of high-performance non-standard hydraulic cylinders such as typical metallurgy and engineering machinery. The following steps are included:

[0132] (1) With the help of Matlab, C#, database and other simulation optimization tools and development tools, a full-process production scheduling optimization algorithm library for hydraulic cylinders is developed and integrated with the production planning and scheduling module.

[0133] (2) Adopting layered development ideas such as MVC and three-tier architecture, we develop reconfigurable middleware for the system foundation layer, namely database interaction, reporting, permissions, etc., and the business creation layer, namely module creation, form configuration, and business constraints.

[0134] (3) Based on the existing discrete manufacturing PLM / ERP / MES integrated platform and the full-process business data of hydraulic cylinder process production and assembly, combined with the system architecture and reconfigurable key technologies constructed above, a high-performance non-standard hydraulic cylinder full-process PLM / ERP / MES integrated optimization platform was developed.

[0135] (4) The full-process production of high-performance non-standard hydraulic cylinders in typical metallurgy, forging, casting, engineering machinery, etc. is applied to verify the feasibility and effectiveness of the full-process production scheduling optimization method of hydraulic cylinders, and the full-process production scheduling optimization algorithm library and integrated optimization platform system functions of non-standard hydraulic cylinders are continuously improved and perfected, so that the above results can be better applied and promoted in actual production management and scheduling optimization.

[0136] The present invention aims to solve the problem that the important basic data of non-standard hydraulic cylinder production scheduling, such as the labor quota, is difficult to obtain accurately due to the complex coupling relationship between process design and manufacturing resources. Based on the digital product geometry and machining process design source, the coupling mechanism between product process design and manufacturing resources is explored innovatively, thereby establishing a machining labor quota model. This model deeply combines process design with manufacturing resources, and can provide more original and more accurate basic data for the fundamental indicator of minimizing the maximum completion time of production scheduling, thereby improving the scientificity and practicality of the scheduling optimization model.

[0137] There are many categories of high-performance non-standard hydraulic cylinders, such as bending cylinders, locking cylinders, AGC cylinders, long-stroke internal guide cylinders, etc. The assembly digital model assemblies of different categories of hydraulic cylinders are very different, and the cylinder diameters of the same category are different, resulting in large differences in assembly processes; each set of hydraulic cylinder assembly involves positioning, clamping, connection, sealing and other actions of each sub-component, and the whole process is discrete manual or semi-automatic; the assembly process card has a large amount of data, complex digital model structure, many professional terms, and difficult to extract process information, resulting in problems such as difficult to accurately assemble the working time. The present invention innovatively proposes a method of first classification and then prediction. First, based on the hydraulic cylinder assembly process data, model and its big data characteristics, a fast classification method for non-standard hydraulic cylinder assembly processes based on process digital model mining is explored. Based on the classification results, an assembly time prediction model based on deep learning is built, and finally an assembly time quota model is established. This research idea not only follows the principle of large classification design of discrete manufacturing product processes, but also takes into account the nonlinear method of artificial intelligence big data prediction. This research idea can be used as a reference for other fields such as process, production, and quality of discrete manufacturing of complex products with strong experience and weak design.

[0138] Finally, the full-process production of high-performance non-standard hydraulic cylinders is a long-term multi-stage production process with a mixture of process design, machining, and assembly discrete processes, as well as the complex coupling relationship between process design and manufacturing resources. This invention proposes for the first time a full-process production scheduling optimization framework that rationally plans and organically integrates the full-process production scheduling problems, methods, and technologies of non-standard hydraulic cylinders: a comprehensive and in-depth study of the full-process production mode of hydraulic cylinders, the division of production stages, the process and production characteristics of each stage, and mutual mapping with workshop scheduling theories and methods, proposes specific production scheduling optimization methods for each production stage, establishes a full-process production scheduling optimization framework for hydraulic cylinders, and then conducts research on scheduling optimization methods for each stage of the full-process production of hydraulic cylinders, in order to improve the overall and optimization effects of hydraulic cylinder production scheduling. At the same time, this research idea, method, and some results are expected to be used for production scheduling in the field of long-term multi-stage production modes in other manufacturing industries.

[0139] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. Only the preferred embodiments of the present invention are expressed. The description is more specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. As long as there is no contradiction in the combination of these technical features, they should be considered as the scope recorded in this specification.

[0140] It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these modifications and improvements all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A full-process production scheduling method for high-performance non-standard hydraulic cylinders, characterized in that: The following steps are involved: Step S1: Obtain the characteristics of each part and processing equipment, build a machining time prediction model, and predict the machining time of each part, including the following steps: Step S11: the characteristics of the equipment include cutting speed, feed rate and cutting depth, and the characteristics of the part include workpiece size and material hardness; Step S12: training a neural network model based on historical data of equipment characteristics and part characteristics and historical data of machining hours; Step S13: Obtain the characteristic data of the parts to be processed and the available equipment, input the trained neural network model, and obtain the predicted machining time of each part; Step S2: Classify the assembly process of high-performance non-standard hydraulic cylinders according to the information in the assembly process card of the hydraulic cylinder, and use the proximal strategy optimization algorithm to predict the assembly time of various types of high-performance non-standard hydraulic cylinders according to the predicted machining hours of each part, including the following steps: Step S211: Initializing the parameters of the proximal strategy optimization algorithm and the environmental parameters, wherein the environmental parameters include the number of workpieces, the number of machines, the process and processing time of each workpiece; Step S212: obtaining an initial state, inputting the initial state into a policy network, outputting a probability distribution of selecting each workpiece, and selecting an action from the action space according to the probability distribution; Step S213: Execute the selected action, update the state matrix and instant reward; Step S214: storing the current state, action, reward and new state into the experience playback buffer; Step S215: Calculate the discounted cumulative return for each time step using the stored rewards, and update the parameters of the policy network using the discounted cumulative return and the probability distribution of the policy network output through the policy gradient method; Update the parameters of the value network by minimizing the prediction error of the state; Step S216: Repeat steps S212 to S215 until the set maximum number of iterations is reached; Step S3: Establish a multi-objective optimization function including minimization of assembly time and minimization of processing cost, solve the multi-objective optimization function, and obtain an optimal production scheduling plan for high-performance non-standard hydraulic cylinders.

2. The full-process production scheduling method for high-performance non-standard hydraulic cylinders according to claim 1 is characterized in that: In step S1, the predicted machining time is optimized by the following expression: In the formula, H i is the machining time of optimized part i; Tmq i is the machining time of part i; λ is the learning index; S(j,i) is the similarity between part j and part i; β is the control parameter of truncated learning effect; n is the total number of similar features; d k is the weight of similar feature i; r k (i, j) is the similarity value between parts i and j under similar feature k.

3. The full-process production scheduling method for high-performance non-standard hydraulic cylinders according to claim 1 is characterized in that: The step S2 classifies the assembly process of the high-performance non-standard hydraulic cylinder according to the information in the assembly process card of the hydraulic cylinder, including the following steps: Step S201: obtaining text data in a plurality of assembly process cards as a training data set, performing text segmentation on the training data set, and removing stop words in the training data set after segmentation to obtain segmentation data; Step S202: Calculate the feature value of each word in the word segmentation data, and select words whose feature value is greater than a set feature value threshold to form a feature word library; Step S203: Calculate the frequency of each category of assembly process cards in the training data set as the prior probability of the category; for each category, calculate the conditional probability of each feature word appearing in the category; Step S204: according to the vocabulary in the feature word library, the posterior probability that the assembly process card to be classified belongs to each category is calculated using the Bayesian formula, and the category with the highest posterior probability is selected as the predicted category of the assembly process card to be classified.

4. The full-process production scheduling method for high-performance non-standard hydraulic cylinders according to claim 1 is characterized in that: The expression for minimizing the assembly time in step S3 is: minC max =max(E p ,p=1,2,...,|p); E p =S p +t aqp ; In the formula, C max is the assembly completion time of the last product; E p is the assembly completion time of product p, S p is the assembly start time of product p; t aqp is the predicted assembly time of product p; L p is the latest arrival time of all workpieces that make up product p; Z p,p' is a decision variable. If all parts of product p arrive at the factory, the previous product p′ has not been assembled, then it is 1; otherwise, it is 0. p′ is the assembly completion time of the immediate predecessor product p′ of product p; μ j,p is a decision variable, which is 1 if workpiece j belongs to product p, otherwise it is 0; θ j,h is a decision variable, which is 1 if product p is transported by vehicle h, otherwise it is 0; S h is the start transportation time of vehicle h; η j,f is a decision variable, which is 1 if product p is produced in factory f, otherwise it is 0; f is the transportation time from factory f to the assembly plant; E j is the processing completion time of workpiece j.

5. The full-process production scheduling method for high-performance non-standard hydraulic cylinders according to claim 1 is characterized in that: The expression for minimizing the processing cost in step S3 is: minTC=TPC+TICj+TAC+TICp+TDC+TTC; Where TPC is the cost of the processing process; F is the total number of factories; M is the total number of machines in factory f; n is the total number of workpieces that need to be processed by machine m; r j is the total number of processes for workpiece j; k j is the processing cost per unit time; x j,i,f,m is a decision variable, which is 1 if the i-th process of workpiece j is processed on the m-th machine in factory f, otherwise it is 0; t j,i,f,m is the time required for the mth machine in factory f to process the i-th process of workpiece j; TICj is the inventory cost of the workpiece; N is the total number of products; v j is the unit time inventory of workpieces; μ j,p is a decision variable, which is 1 if workpiece j belongs to product p, otherwise it is 0; Sa p is the assembly start time of product p; E j is the processing completion time of workpiece j; TAC is the cost of the assembly process; k p is the assembly cost per unit time; t p is the assembly time of product p; TICp is the inventory cost of the product; H is the total number of vehicles; v p is the unit time inventory cost of the product; σ p,h is a decision variable, which is 1 if product p is assigned to vehicle h, otherwise it is 0; Ea p is the assembly completion time; TDC is the transportation cost; ω is the fixed cost of the vehicle; U f,c A collection of factories and customers' locations; k h is the transportation cost per unit time; is a decision variable, which is 1 if d is a destination in the delivery process of vehicle h, otherwise it is 0; d,d' is a decision variable, which is 1 if location d' is delivered immediately after location d, otherwise it is 0; t d,d' is the transportation time from location d' to location d; TTC is the delay cost; C is the total number of customers; k T is the delay cost per unit time; p,c is a decision variable, which is 1 if product p belongs to customer c, otherwise it is 0; T c The time it takes for a vehicle to deliver the product to customer c; E c is the delivery deadline for customer c.

6. A full-process production scheduling system for high-performance non-standard hydraulic cylinders, based on a full-process production scheduling method for high-performance non-standard hydraulic cylinders as claimed in any one of claims 1 to 5, characterized in that: It includes machining time prediction module, assembly time prediction module and production scheduling optimization module; The machining time prediction module is used to train a machining time prediction model based on historical machining data, and input the machining data of the workpiece to be predicted into the trained machining time prediction model to obtain the machining time of the workpiece to be predicted; The assembly time prediction module: trains an assembly time prediction model based on historical assembly data, classifies various assembly tasks of high-performance non-standard hydraulic cylinders, inputs the classified assembly data into the trained assembly time prediction model, and obtains the assembly time of each assembly task; The production scheduling optimization module formulates a production scheduling plan based on the predicted time of machining and assembly, and dynamically adjusts the production scheduling plan based on real-time production data and external changes.

7. The full-process production scheduling system for high-performance non-standard hydraulic cylinders according to claim 6 is characterized by: The machining time prediction module also includes a machining time prediction optimization module, and the machining time prediction optimization module optimizes the machining time prediction value of each part by calculating the similarity between the parts.

8. The full-process production scheduling system for high-performance non-standard hydraulic cylinders according to claim 7 is characterized in that: The assembly time prediction module includes an assembly process classification module, which constructs a feature word library by extracting feature values ​​of text data in a plurality of assembly process cards, and classifies the assembly process cards using a Bayesian algorithm based on the feature word library.

Images