Data-driven production capacity prediction method and apparatus for automated production line

Abstract

A data-driven production capacity prediction method and apparatus for an automated production line, relating to the technical field of data processing. On the basis of various types of historical feature data of a surface-mount production line device, key features affecting production capacity are extracted via mutual information, and a PSO-ELM prediction model is constructed, thereby enabling intelligent prediction of production line capacity, and laying a foundation for refined and intelligent scheduling optimization and production capacity planning. The present invention fully takes into account dynamic disturbance event data to correct the prediction model, thereby effectively improving the accuracy of the prediction model.

Description

Data-driven automated production line capacity prediction method and device TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a data-driven automated production line capacity prediction method and device. BACKGROUND

[0002] Automated production lines, such as surface mount technology production lines, have many types of equipment, complex processes, many uncertain factors, and many work-in-process products. The production characteristics are multi-variety, variable batch, batch-research mixed production, many dynamic disturbance events, short delivery cycle, and many precise consumables for chip mounter equipment. The above characteristics make it difficult to predict capacity.

[0003] Existing data-driven machine learning production line capacity prediction assumes that all equipment on the production line is running normally during the production cycle or considers the time of equipment failure or downtime that has occurred, ignoring the impact of dynamic disturbance events on production line capacity during the production cycle. For example, important equipment (chip mounter) critical component consumables dynamic maintenance time (such as replacing spare parts when the preset of the mounter feeder exceeds 1.1 million points, replacing spare parts when the suction nozzle exceeds the offline time), and potential failures (such as the stability of the mounter decreases after 3-5 years of use, which can easily lead to a non-compliant CPK of the mounter capacity index).

[0004] Therefore, how to fully integrate equipment health management and mine automated production line manufacturing process data to achieve capacity prediction is one of the problems to be solved. SUMMARY

[0005] (I) Technical problems solved

[0006] To solve the problems of the prior art, the present application provides a data-driven automated production line capacity prediction method and device, which solves the problem that the existing capacity prediction method ignores the impact of dynamic disturbance events on production line capacity.

[0007] (II) Technical solutions

[0008] To achieve the above purpose, the present application is implemented by the following technical solutions:

[0009] In a first aspect, an automated production line capacity prediction method is provided, which comprises:

[0010] Obtaining a plurality of historical feature data and a plurality of current feature data of the manufacturing process of the automated production line, and preprocessing them;

[0011] Obtaining mutual information values of each preprocessed historical feature data and order processing time, and screening out several key feature data according to the mutual information values;

[0012] A sample dataset for training the capacity prediction model is constructed based on several key feature data, and the capacity prediction model is trained based on an optimization algorithm to obtain a trained capacity prediction model.

[0013] The training process includes the following steps:

[0014] Based on various current characteristic data of the automated production line manufacturing process, it is determined whether a dynamic disturbance event occurred during the production cycle. If not, the original predicted value output by the capacity prediction model is used as the final predicted value. If the event occurred, the original predicted value y before the dynamic disturbance event occurred is obtained. pre (t) and actual value y rea (t), and calculate the first residual value er(t)=y rea (t)-y pre (t); Obtain the original predicted value y after a dynamic disturbance event occurs. pre (T) and actual value y rea (T), and calculate the second residual value er(T)=y rea (T)-y pre (T); Determine whether er(t) and er(T) simultaneously meet the tolerance requirements. If they do, then set y pre (t) or y pre Any one of (T) is used as the final predicted value; otherwise, the residuals er(T) corresponding to the n time points before the dynamic disturbance event are obtained. i ), n = 1, 2, ..., n; n is the number of categories of key feature data to be screened; and it is used as the input variable of the capacity prediction model to obtain the predicted compensation value y. err (T); and the predicted compensation value y err (T) and the original predicted value y pre Adding (T) together yields the corrected prediction value y. final (T) and used as the final predicted value;

[0015] Input the key feature data to be predicted into the capacity prediction model to obtain the original predicted value corresponding to the key feature data. If no dynamic disturbance event occurs during the production cycle or the tolerance requirement is met, the original predicted value output by the capacity prediction model is used as the final predicted value; otherwise, the corrected predicted value y corresponding to the key feature data is used. final (T) is used as the final predicted value.

[0016] Furthermore, the preprocessing includes missing value completion and data normalization.

[0017] Further, the capacity prediction model is a prediction model based on a limit learning machine, and the capacity prediction model is optimized by using a particle swarm optimization algorithm.

[0018] and the capacity prediction model is:

[0019] wherein,

[0020] x t represents an input vector of a capacity prediction model;

[0021] β i represents an output weight, and there are L of them;

[0022] g() represents an activation function;

[0023] ω i represents an input weight, and there are L of them;

[0024] b i represents a bias of a hidden layer node;

[0025] y t represents x t , and the corresponding output vector.

[0026] Further, the objective function is used as the fitness function of the particle swarm optimization algorithm; and the objective function f: min f = || βH-Y || 2 β = H + Y

[0027] wherein, β represents an output weight matrix, β = [β1, β2, …, β i ,…β L ]; β i represents an output weight of the capacity prediction model, and there are L of them; b i represents a bias of a hidden layer node, and there are L of them; i = 1, 2, …, L;

[0028] x tn represents the corresponding nth key feature data;

[0029] H represents a matrix;

[0030] H + is the Moore-Penrose generalized inverse matrix of the matrix H;

[0031] Y represents a true value.

[0032] Further, the updating method of the speed and position in the particle swarm optimization algorithm is:

[0033] wherein, respectively represent the motion speed of the i-th particle in the k-th and k+1-th iteration;

[0034] ω is an inertia weight;

[0035] c1, c2 are learning factors;

[0036] r1, r2 are random numbers between 0 and 1;

[0037] is the historical optimal position of the i-th particle in the k-th iteration;

[0038] is the historical optimal position of the population in the k-th iteration;

[0039] respectively represent the position of the i-th particle in the k-th and k+1-th iteration.

[0040] Further, the calculation method of the inertia weight ω is:

[0041] wherein f i represents the objective function value of the i-th particle;

[0042] f min represents the minimum value in the historical value of the objective function;

[0043] f avg represents the average value of the historical value of the objective function;

[0044] ω max represents the maximum weight; ω min represents the minimum weight.

[0045] Further, during training, the mean absolute percentage error, the mean absolute error, and the normalized mean square error are used as evaluation criteria for the capacity prediction model.

[0046] In a second aspect, an automatic production line capacity prediction device is provided, and the device comprises:

[0047] a data acquisition and preprocessing module, configured to acquire a plurality of historical characteristic data and a plurality of current characteristic data of a manufacturing process of an automatic production line, and to preprocess the data;

[0048] a key characteristic data screening module, configured to acquire mutual information values of each preprocessed historical characteristic data and processing time of an order, and to screen a plurality of key characteristic data according to the mutual information values;

[0049] The capacity prediction model training module is used to construct a sample dataset for training the capacity prediction model based on several key feature data, and to train the capacity prediction model based on an optimization algorithm to obtain a trained capacity prediction model.

[0050] The training process includes the following steps:

[0051] Based on various current characteristic data of the automated production line manufacturing process, it is determined whether a dynamic disturbance event occurred during the production cycle. If not, the original predicted value output by the capacity prediction model is used as the final predicted value. If the event occurred, the original predicted value y before the dynamic disturbance event occurred is obtained. pre (t) and actual value y rea (t), and calculate the first residual value er(t)=y rea (t)-y pre (t); Obtain the original predicted value y after a dynamic disturbance event occurs. pre (T) and actual value y rea (T), and calculate the second residual value er(T)=y rea (T)-y pre (T); Determine whether er(t) and er(T) simultaneously meet the tolerance requirements. If they do, then set y pre (t) or y pre Any one of (T) is used as the final predicted value; otherwise, the residuals er(T) corresponding to the n time points before the dynamic disturbance event are obtained. i ), n = 1, 2, ..., n; n is the number of categories of key feature data to be screened; and it is used as the input variable of the capacity prediction model to obtain the predicted compensation value y. err (T); and the predicted compensation value y err (T) and the original predicted value y pre Adding (T) together yields the corrected prediction value y. final (T) and used as the final predicted value;

[0052] The prediction result output module is used to input the key feature data to be predicted into the capacity prediction model and obtain the original predicted value corresponding to the key feature data. If no dynamic disturbance event occurs during the production cycle or the tolerance requirement is met, the original predicted value output by the capacity prediction model is used as the final predicted value; otherwise, the corrected predicted value y corresponding to the key feature data is used. final (T) is used as the final predicted value.

[0053] Thirdly, a computer-readable storage medium is provided that stores a computer program for predicting the capacity of an automated production line, wherein the computer program causes a computer to execute the aforementioned method for predicting the capacity of an automated production line.

[0054] In a fourth aspect, an electronic device is provided, comprising:

[0055] one or more processors;

[0056] a memory; and

[0057] one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, characterized in that the programs comprise a program for executing the above-mentioned automated production line capacity prediction method.

[0058] (III) Beneficial Effects

[0059] The present application provides an automated production line capacity prediction method and device based on data driving. Compared with the prior art, the following beneficial effects are possessed:

[0060] 1) Traditional capacity evaluation mostly adopts artificial experience, simulation modeling method and optimization method, wherein the artificial experience evaluation has low efficiency and low accuracy, the modeling method is difficult to construct a model for different production business scenarios, and is difficult to fully reflect the time-varying characteristics, complexity and uncertainty of the actual production state, and has low prediction efficiency and poor generalization performance. The optimization method is based on NP-hard problem mathematical modeling under different constraint conditions, and has low solving efficiency and certain limitations in solving quality. Moreover, the input data of the above-mentioned model is mostly static, offline and small amount of data, and the method is difficult to adapt to the dynamic and high-dimensional data of the surface mounting production line. The present application is based on historical characteristic data such as the running state of the surface mounting line equipment, the state duration, the number of PCB boards, the geometric data, the order quantity and the priority, extracts key features affecting the capacity through mutual information, constructs a PSO-ELM prediction model, realizes intelligent prediction of the production line capacity, and lays a foundation for fine and intelligent scheduling optimization and capacity planning.

[0061] 2) The existing data-driven production line capacity prediction model mostly considers the device failure time or downtime that occurs after the event, ignores the unnecessary disturbance or even downtime of the production line caused by the dynamic maintenance time of the mounting equipment key consumables and potential faults (such as the non-compliance of the machine capacity index CPK) on the production line during the production cycle, and leads to inaccurate prediction model accuracy, which is difficult to provide a scientific basis for the overall efficiency improvement of the production line equipment, fine control of the manufacturing process, and dynamic allocation of manufacturing resources. The present application fully considers the dynamic maintenance time of the mounting equipment consumables (such as the preset of the mounting equipment feeder exceeding 1.1 million points to automatically lock the code for replacement of spare parts, the suction nozzle exceeding the offline time to replace spare parts, and the replacement head module replacement time) and the monitoring machine CPK value change trend, and the above-mentioned data is used as dynamic disturbance data to correct the prediction model, effectively improving the prediction model accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0062] The technical solutions in the embodiments of the present application or the prior art will be clearly and completely described below with reference to the accompanying drawings that need to be used in the embodiments or the prior art description. Obviously, the accompanying drawings below are only some embodiments of the present application, and other drawings can be obtained by a person of ordinary skill in the art without creative effort.

[0063] FIG. 1 is a flowchart of an embodiment of the present application;

[0064] FIG. 2 is a flowchart of prediction value correction of an embodiment of the present application. DETAILED DESCRIPTION

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application are described clearly and completely. Obviously, the described embodiments are only some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative effort fall within the protection scope of the present application.

[0066] The embodiments of the present application provide a data-driven-based automated production line capacity prediction method and device, which solve the problem that the existing capacity prediction method ignores the influence of dynamic disturbance events on the production line capacity.

[0067] To better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the accompanying drawings and specific embodiments of the specification.

[0068] Embodiment 1

[0069] As shown in FIG. 1, the present application provides an automated production line capacity prediction method, which is executed by a computer, and the method comprises the following steps:

[0070] Obtaining a plurality of historical feature data and a plurality of current feature data of the manufacturing process of the automated production line, and preprocessing the historical feature data and the current feature data;

[0071] Obtaining mutual information values of each preprocessed historical feature data and processing time of an order, and screening a plurality of key feature data according to the mutual information values;

[0072] Constructing a sample data set for training a capacity prediction model based on the plurality of key feature data, and training the capacity prediction model based on an optimization algorithm to obtain a trained capacity prediction model;

[0073] In the training, the following steps are included:

[0074] Based on the current feature data of the automated production line manufacturing process, it is determined whether a dynamic disturbance event occurs within a production cycle; if not, the original prediction value output by the production capacity prediction model is taken as the final prediction value, and if so, the original prediction value y pre (t) before the dynamic disturbance event occurs is obtained rea (t) is calculated to obtain a first residual value er(t) = y rea (t) - y pre (t); the original prediction value y pre (T) after the dynamic disturbance event occurs is obtained rea (T) is calculated to obtain a second residual value er(T) = y rea (T) - y pre (T); it is determined whether er(t) and er(T) satisfy the tolerance requirement at the same time, if so, any one of y pre (t) or y pre (T) is taken as the final prediction value; otherwise, the residual er(T i ) corresponding to the n time points before the dynamic disturbance event occurs is obtained, n = 1, 2, …, n; n is the number of types of key feature data for screening; and is taken as an input variable of the production capacity prediction model to obtain a prediction compensation value y err (T); the prediction compensation value y err (T) is added to the original prediction value y pre (T) to obtain a corrected prediction value y final (T) and taken as the final prediction value.

[0075] The key feature data to be predicted is input into the production capacity prediction model to obtain an original prediction value corresponding to the key feature data to be predicted, if no dynamic disturbance event occurs within the production cycle or the tolerance requirement is met, the original prediction value output by the production capacity prediction model is taken as the final prediction value, otherwise, the corrected prediction value y final (T) corresponding to the key feature data to be predicted is taken as the final prediction value.

[0076] The beneficial effects of the embodiment are:

[0077] 1) The traditional production capacity evaluation method is difficult to adapt to the characteristics of dynamic and high-dimensional data of the surface mount production line. The present application is based on the multiple historical feature data of the surface mount line equipment, extracts the key features affecting the production capacity through mutual information, constructs a PSO-ELM prediction model, realizes intelligent prediction of the production line capacity, and lays a foundation for fine and intelligent scheduling optimization and production capacity planning.

[0078] 2) The existing data-driven production line capacity prediction model mainly considers the time of equipment failure or downtime after the event, ignores the unnecessary disturbance or even shutdown of the production line caused by dynamic disturbance event data, and leads to inaccurate prediction model accuracy, which is difficult to provide scientific basis for the overall efficiency improvement of the production line equipment, fine control of the manufacturing process, and dynamic allocation of manufacturing resources. The present application fully considers the dynamic disturbance event data to correct the prediction model, effectively improving the accuracy of the prediction model.

[0079] The following will take the surface mount production line as an example, which has the characteristics of multi-variety, variable batch, and batch-research mixed production. The detailed implementation process of the embodiment of the present application will be described:

[0080] S1, obtain a plurality of historical feature data and a plurality of current feature data of the surface mount production line manufacturing process, and pre-process them;

[0081] In specific implementation, the feature data of the surface mount production line manufacturing process includes device, product, maintenance, order and other related feature data;

[0082] Among the above-mentioned four types of feature data of device, product, maintenance and order, according to their time attributes, they can be further divided into historical feature data and current feature data.

[0083] Among them, the historical feature data can be directly called by reading the data in the database, mainly used for training the original prediction model, such as reading the device static attribute information data (device service life, chip mounter type, device process type) in the setting period, the delivered similar product data, the regular maintenance time and frequency of each process equipment, and the completed similar product order.

[0084] The current feature data is mainly used for correcting the original prediction model to improve its prediction accuracy and robustness, which can be obtained by real-time reading of device work log or reading of existing information system of enterprise. The current feature data mainly includes the current working state data of each process equipment, the dynamic maintenance time of the equipment, the sudden abnormal processing time, and the current product data, the current order data, etc.

[0085] The device-related feature data of the surface mount production line manufacturing process mainly includes: state, attribute information data of each process equipment; for example: running state data (such as device working, idle, waiting time, etc.) of plate loading machine, printing machine, chip mounter, reflow soldering, SPI detection equipment, X-ray, etc., equipment service life data (such as 1-3 years, 4-6 years, more than 7 years), chip mounter type data (high-speed chip mounter, medium-speed chip mounter, low-speed chip mounter), equipment process data (mounting equipment module quantity, module type). The above data can be obtained by reading the device working log through socket, existing information management system or device management record form.

[0086] The product-related feature data mainly includes: PCB board length, width, thickness geometric parameters, PCB board quantity, PCB weight, single board point number, PCB process condition (single-sided or double-sided), and piece number. The above data can be obtained from product BOM and record form.

[0087] The maintenance-related feature data mainly includes: spare parts quantity of each process equipment, maintenance time, spare parts replacement time, maintenance frequency, fault frequency, repair time, etc. For example: heavy part fault data of chip mounter (X / Y axis wear, PCB transmission anomaly, clamping plate mechanism air pressure anomaly), heavy part maintenance data of chip mounter (supplying device production part count, replacement head, nozzle up and down line time, machine CPK value). The above data can be obtained from information management system, device software system, and mounting equipment error-proofing part system.

[0088] The order-related feature data mainly includes: order quantity, order priority, production scheduling, order information, etc. The above data is obtained through ERP / MES system.

[0089] The preprocessing includes: missing value completion and data normalization.

[0090] The missing value completion includes: extracting time, numerical value and Boolean value of the feature data of each process equipment, product, maintenance and order related to the surface mount production line in a production cycle, converting the time format into numerical value, setting 0 for False and 1 for True, completing the missing value of the above data, setting 0 for NAN, and using mean interpolation for the missing part of the remaining non-empty value.

[0091] In order to prevent the difference in numerical dimension from affecting the prediction accuracy during production manufacturing process data mining, the maximum-minimum method is used for normalization before data mining. All data is mapped to between 0 and 1, that is,

[0092] Among them,

[0093] x inorm is the normalized data;

[0094] x ij is the original data to be processed;

[0095] x imin is the minimum value in the original data; x imax is the maximum value in the original data, set by manual.

[0096] S2, obtain the mutual information value of each pre-processed historical feature data and the processing time of the order, and filter out several key feature data according to the mutual information value;

[0097] Since there are linear correlation and nonlinear correlation between the data of each process equipment, product data, order data and maintenance data on the surface mount production line, a single linear correlation and nonlinear correlation analysis method is difficult to solve. Therefore, the mutual information method is used to extract the key features affecting the production capacity in the embodiment;

[0098] The production capacity of the surface mount production line is mainly measured by the processing time (cycle time, CT) of the PCB order on the production line. Cycle Time refers to the production rhythm, i.e. running time / actual production quantity;

[0099] In specific implementation, the mutual information value between various feature data and the processing time CT of the order is calculated according to the formula, and the mutual information between CT and each feature parameter is analyzed;

[0100] wherein I(X;Y) represents the mutual information value between different features and CT;

[0101] p(x,y) is the joint probability density function;

[0102] p(x) and p(y) are the marginal probability density functions of x and y, respectively;

[0103] p(x) is the probability of occurrence of each feature x element in the feature set X;

[0104] p(y) is the probability of occurrence of each feature y element in the feature set Y;

[0105] p(x,y) is the probability of simultaneous occurrence of features x and y elements;

[0106] The above numerical values can be realized by calling the mutual information function in the Python software Sklearn library.

[0107] Example: The feature set X and the feature set Y (output vector set) are shown in Table 1 (all data are normalized by S1 step).

[0108] Table 1

[0109] In the table 1, the order contains processing tasks of four products, and the data in the first row of the table is the equipment state, the three-dimensional value of the PCB, the quantity and the corresponding CT value in the production process of the first specification product in the order; the data in the second row is the equipment state, the three-dimensional value of the PCB, the quantity and the corresponding CT value in the production process of the first specification product in the order, and so on.

[0110] For example, the feature set X and the corresponding feature set Y corresponding to the equipment state are respectively: X=[0.4, 0.2, 0.2, 0.5] Y=[0.03, 0.21, 0.29, 0.21]

[0111] When x=0.4 and y=0.29, it is calculated that: p(x)=0.25 p(y)=0.25 p(x,y)=0.06

[0112] Substituting the mutual information calculation formula, the mutual information I(X;Y) between the equipment state feature and CT can be obtained, which is 0.69. In this way, the mutual information values between other features and CT can be obtained.

[0113] And according to the mutual information values, a plurality of key feature data are screened out, including:

[0114] According to the mutual information values, the various features are sorted, and the corresponding features of those below a set threshold value (for example, the threshold value is set to 0.1) are deleted, and the features corresponding to the mutual information values exceeding the threshold value are reserved as key features to obtain the key feature data. (The specific setting of the threshold value can be adjusted according to the actual working condition).

[0115] S3, based on a plurality of key feature data, a sample data set for training a production capacity prediction model is constructed, and the production capacity prediction model is trained based on an optimization algorithm to obtain a trained production capacity prediction model;

[0116] In specific implementation, it is assumed that a training set consisting of N training samples is represented as:

[0117] Wherein, x t represents a sample input vector; x t =[x t1 ,x t2 ,…x tn ] T ; the vector is an n-dimensional data corresponding to n key feature data;

[0118] y t represents a corresponding output vector; yt = [y t1 ,y t2 ,…y tk ] T ; k represents the number of product categories in the order;

[0119] Example: Assuming that after S2 screening, 5 key feature data affecting production capacity are obtained, and they are: equipment state, PCB length, PCB width, PCB thickness, and PCB quantity; a training sample x t may be represented as: x t = [x t1 ,x t2 ,x t3 ,x t4 ,x t5 ] T

[0120] wherein x t1 represents the equipment state data of all specifications of products in the order, and according to Table 1, it can be represented as: x t1 = {0.4, 0.2, 0.2, 0.5};

[0121] x t2 represents the PCB length data of all specifications of products in the order, and according to Table 1, it can be represented as: x t2 = {0.37, 0.47, 0.22, 0.26}; x t3 ,x t4 ,x t5 and so on;

[0122] and in the multi-variety, variable-batch production mode, different PCBs and different mounting point numbers correspond to different CT values; y t = [y t1 ,y t2 ,y t3 ,y t4 ] T

[0123] wherein y t1 ~ y t4 represent the CT values of the four products, and according to Table 1, they can be represented as: y t = [0.03, 0.21, 0.29, 0.21].

[0124] Due to the production characteristics of the surface mounting production line being multi-variety, variable-batch, and batch-research mixed production, therefore:

[0125] for the mature products that have been batch-produced, the CT values of the historical data of the same type of products in the database are called,

[0126] For new product development, small batch products, manual estimation of experience value or simulator to calculate CT value,

[0127] For large batch products, estimate the CT value after trial production.

[0128] y t Corresponding to the CT value of different products.

[0129] 67% of the sample data as training data set, 33% as test data set; The capacity prediction model is an Extreme Learning Machine (ELM) based prediction model, which can be represented as:

[0130] Wherein,

[0131] β i Indicates the output weight, a total of L;

[0132] g() represents the activation function;

[0133] ω i Indicates the input weight, a total of L;

[0134] b i Indicates the bias of the hidden layer node;

[0135] When training the capacity prediction model, the goal is to minimize the output error. Assuming the error is zero, there are β i , ω i and b i Make the predicted output equal to the true value, minimize the approximate square difference method to solve β, you can set the objective function f as: min f = || βH-Y || 2 β = H + Y

[0136] Wherein, β represents the output weight matrix, β = [β1, β2, β3, … β L ];

[0137] H represents the preset value, which is determined by β i , ω i and b i ;

[0138] H + is the Moore-Penrose generalized inverse matrix of matrix H;

[0139] Y represents the true value;

[0140] The embodiment adopts improved particle swarm optimization (PSO) to optimize input weights and biases of hidden layer nodes of the model, and takes a target function as a fitness function;

[0141] Specifically, the updating method of the speed and the position in the algorithm is as follows:

[0142] wherein, are respectively the motion speeds of the i th particle in the k th and k+1 th iterations;

[0143] ω is an inertia weight;

[0144] c1 and c2 are learning factors;

[0145] r1 and r2 are random numbers generated between 0 and 1;

[0146] is the historical optimal position of the i th particle in the k th iteration;

[0147] is the historical optimal position of the group in the k th iteration;

[0148] are respectively the positions of the i th particle in the k th and k+1 th iterations;

[0149] Further, in the embodiment, in order to avoid falling into a local optimum, the difference of the inertia weight of different particles is ignored, which may cause a good particle to fly away from the optimal position due to the inertia weight being too large. A dynamic adjustment inertia weight method is adopted, and the main idea is to adjust the inertia weight by comparing the fitness value of the current particle with the average fitness value in the population.

[0150] wherein, f i represents a target function value of the i th particle;

[0151] f min represents the minimum value in the historical value of the target function;

[0152] f avg represents the average value of the historical value of the target function;

[0153] ω max represents the maximum weight; ω min represents the minimum weight, which is artificially preset.

[0154] So far, the embodiment completes the construction of the PSO-ELM prediction model. However, the model fails to consider the influence of dynamic disturbance events occurring in the production cycle on the capacity prediction, and the output prediction value needs to be corrected.

[0155] Wherein, as shown in FIG. 2, during training, the following steps are further included:

[0156] Step 1, based on the multiple current feature data of the automated production line manufacturing process, it is judged whether a dynamic disturbance event occurring in the production cycle occurs;

[0157] If not, the original prediction value output by the capacity prediction model is taken as the final prediction value;

[0158] If so, the original prediction value y pre (t) before the dynamic disturbance event occurs is obtained, and the actual value y rea (t) is obtained, and the first residual value er(t) is calculated; and the original prediction value y pre (T) after the dynamic disturbance event occurs is obtained, and the actual value y rea (T) is obtained, and the second residual value er(T) is calculated.

[0159] In specific implementation, the dynamic disturbance event is an event that will affect the capacity, for example:

[0160] The number of 1.1 million parts produced by the mounting equipment feeder is the threshold value, and when the threshold value is exceeded in the production cycle, the dynamic disturbance event occurs, and the replacement of spare parts causes additional time Δt1;

[0161] When the suction nozzle on-off time exceeds the threshold value of 36h, the dynamic disturbance event occurs, and the replacement and maintenance of spare parts cause additional time Δt2;

[0162] When the machine CPK value is not up to standard, the dynamic disturbance event occurs, affecting the mounting offset and mounting head, and additional fault time Δt3 is required;

[0163] Similarly, when other similar dynamic disturbance events occur, additional time will be caused.

[0164] The calculation method of the first residual value er(t) is: er(t) = y rea (t) - y pre (t)

[0165] The calculation method of the second residual value er(T) is: er(T) = y rea (T) - y pre (T) T = t + Δt i

[0166] Wherein, time t represents the time before the dynamic disturbance event occurs;

[0167] Time T represents the time after the dynamic disturbance event occurs;

[0168] y pre (t) represents the original prediction value output by the capacity prediction model before the dynamic disturbance event occurs (t time);

[0169] y rea (t) represents the actual CT value at t time before the dynamic disturbance event occurs;

[0170] y pre (T) represents the original prediction value output by the capacity prediction model after the dynamic disturbance event occurs (T time);

[0171] y rea (T) represents the actual CT value at T time before the dynamic disturbance event occurs;

[0172] Δt i represents the additional time caused by the i-th dynamic disturbance event;

[0173] For example, if a dynamic disturbance event occurs on the 15th day in the production cycle, that is, t = 15, the input data of the previous 15 days is input into the capacity prediction model to obtain the original prediction value y pre (15) before the dynamic disturbance event occurs;

[0174] And the additional time caused by the dynamic disturbance event is 3 days, that is, T = 15 + 3 = 18, the input data of the previous 18 days is input into the capacity prediction model to obtain the original prediction value y pre (18) after the dynamic disturbance event occurs;

[0175] In the training process, y rea (t) and y rea (T) can be obtained by reading the device log through Socket or the existing information system of the enterprise;

[0176] Step 2, judge whether er(t) and er(T) meet the tolerance requirements at the same time;

[0177] If they meet the requirements at the same time, any one of y pre (t) or y pre (T) is taken as the final prediction value;

[0178] Otherwise, the residual error er(T i ) corresponding to the n time points before the dynamic disturbance event occurs is obtained, n = 1, 2, …, n; n is the number of types of key feature data selected; and is taken as the input variable of the capacity prediction model to obtain the prediction compensation value yerr (T); and adding the prediction compensation value y err (T) to the original prediction value y pre (T) to obtain a corrected prediction value y final (T) as the final prediction value.

[0179] wherein the residual error er(T i ) is calculated in the same way as er(t) and er(T).

[0180] In this embodiment, during training, the mean absolute percentage error (MAPE), the mean absolute error (MAE), and the normalized mean square error (RMSE) are used as evaluation criteria for the model prediction results.

[0181] S4, input the key feature data to be predicted into the capacity prediction model to obtain an original prediction value corresponding to the key feature data to be predicted. If no dynamic disturbance event occurs within the production cycle or the tolerance requirement is met, the original prediction value output by the capacity prediction model is taken as the final prediction value. Otherwise, the corrected prediction value y final (T) corresponding to the key feature data to be predicted is taken as the final prediction value.

[0182] Embodiment 2:

[0183] An automatic production line capacity prediction device, comprising:

[0184] a data acquisition and preprocessing module configured to acquire a plurality of historical feature data and a plurality of current feature data of an automatic production line manufacturing process and to preprocess the data;

[0185] a key feature data screening module configured to acquire mutual information values of each preprocessed historical feature data and order processing time, and to screen a plurality of key feature data according to the mutual information values;

[0186] a capacity prediction model training module configured to construct a sample data set for training a capacity prediction model based on the plurality of key feature data, and to train the capacity prediction model based on an optimization algorithm to obtain a trained capacity prediction model;

[0187] wherein during training, the following steps are included:

[0188] based on the plurality of current feature data of the automatic production line manufacturing process, determine whether a dynamic disturbance event occurs within the production cycle; if not, the original prediction value output by the capacity prediction model is taken as the final prediction value; if so, the original prediction value y pre (t) before the dynamic disturbance event occurs is obtained, and a first residual error value er(t) = y rea (t) is calculated.rea (t)-y pre (t); obtaining the original predicted value y after the dynamic disturbance event occurs pre (T) and the actual value y rea (T), and calculating a second residual value er(T) = y rea (T)-y pre (T); determining whether er(t) and er(T) satisfy the tolerance requirement at the same time, if they satisfy the tolerance requirement at the same time, y pre (t) or y pre (T) as the final predicted value; otherwise, obtaining the residual er(T i ) corresponding to the n time points before the dynamic disturbance event occurs, n = 1, 2, …, n; n is the number of types of key feature data for screening; and taking the residual er(T err ) as an input variable of the capacity prediction model to obtain a predicted compensation value y err (T); and adding the predicted compensation value y pre (T) and the original predicted value y final (T) to obtain a corrected predicted value y final (T) and taking it as the final predicted value.

[0189] The prediction result output module is configured to input the key feature data to be predicted into the capacity prediction model to obtain an original predicted value corresponding to the key feature data to be predicted, and if no dynamic disturbance event occurs within the production cycle or the tolerance requirement is met, taking the original predicted value output by the capacity prediction model as the final predicted value, otherwise, taking the corrected predicted value y final (T) corresponding to the key feature data to be predicted as the final predicted value.

[0190] Embodiment 3:

[0191] A computer readable storage medium stores a computer program for capacity prediction of an automated production line, wherein the computer program causes a computer to perform the following steps:

[0192] Obtain a plurality of historical feature data and a plurality of current feature data of the manufacturing process of the automated production line, and pre-process the historical feature data and the current feature data;

[0193] Obtain mutual information values of each pre-processed historical feature data and processing time of an order, and screen a plurality of key feature data according to the mutual information values;

[0194] Construct a sample data set for training a capacity prediction model based on the plurality of key feature data, and train the capacity prediction model based on an optimization algorithm to obtain a trained capacity prediction model;

[0195] Wherein, during the training, the following steps are included:

[0196] Based on various current characteristic data of the automated production line manufacturing process, it is determined whether a dynamic disturbance event occurred during the production cycle. If not, the original predicted value output by the capacity prediction model is used as the final predicted value. If the event occurred, the original predicted value y before the dynamic disturbance event occurred is obtained. pre (t) and actual value y rea (t), and calculate the first residual value er(t)=y rea (t)-y pre (t); Obtain the original predicted value y after a dynamic disturbance event occurs. pre (T) and actual value y rea (T), and calculate the second residual value er(T)=y rea (T)-y pre (T); Determine whether er(t) and er(T) simultaneously meet the tolerance requirements. If they do, then set y pre (t) or y pre Any one of (T) is used as the final predicted value; otherwise, the residuals er(T) corresponding to the n time points before the dynamic disturbance event are obtained. i ), n = 1, 2, ..., n; n is the number of categories of key feature data to be screened; and it is used as the input variable of the capacity prediction model to obtain the predicted compensation value y. err (T); and the predicted compensation value y err (T) and the original predicted value y pre Adding (T) together yields the corrected prediction value y. final (T) and used as the final predicted value;

[0197] Input the key feature data to be predicted into the capacity prediction model to obtain the original predicted value corresponding to the key feature data. If no dynamic disturbance event occurs during the production cycle or the tolerance requirement is met, the original predicted value output by the capacity prediction model is used as the final predicted value; otherwise, the corrected predicted value y corresponding to the key feature data is used. final (T is the final predicted value.)

[0198] Example 4:

[0199] An electronic device, comprising:

[0200] One or more processors;

[0201] Memory; and

[0202] One or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the programs including steps for performing the following:

[0203] Obtain a plurality of historical feature data and a plurality of current feature data of an automated production line manufacturing process, and preprocess them;

[0204] Obtain mutual information values of each preprocessed historical feature data and processing time of an order, and screen out a plurality of key feature data according to the mutual information values;

[0205] Construct a sample data set for training a production capacity prediction model based on the plurality of key feature data, and train the production capacity prediction model based on an optimization algorithm to obtain a trained production capacity prediction model;

[0206] During training, the following steps are included:

[0207] Based on a plurality of current feature data of an automated production line manufacturing process, determine whether a dynamic disturbance event occurs within a production cycle; if not, the original prediction value output by the production capacity prediction model is taken as the final prediction value; if so, the original prediction value y pre (t) before the dynamic disturbance event occurs is obtained, and the actual value y rea (t) is calculated to obtain a first residual value er(t) = y rea (t)-y pre (t); the original prediction value y pre (T) after the dynamic disturbance event occurs is obtained, and the actual value y rea (T) is calculated to obtain a second residual value er(T) = y rea (T)-y pre (T); determine whether er(t) and er(T) satisfy the tolerance requirement at the same time, if so, take any one of y pre (t) or y pre (T) as the final prediction value; otherwise, obtain the residual er(T i ) corresponding to the n time points before the dynamic disturbance event occurs, n = 1, 2, …, n; n is the number of types of screened key feature data; and take it as an input variable of the production capacity prediction model to obtain a prediction compensation value y err (T); and add the prediction compensation value y err (T) and the original prediction value y pre (T) to obtain a corrected prediction value y final (T) and take it as the final prediction value;

[0208] Input the key feature data to be predicted into the production capacity prediction model to obtain the original prediction value corresponding to the key feature data to be predicted, if no dynamic disturbance event occurs within the production cycle or the tolerance requirement is met, the original prediction value output by the production capacity prediction model is taken as the final prediction value, otherwise, the corrected prediction value y final (T) corresponding to the key feature data to be predicted is taken as the final prediction value.

[0209] It can be understood that the automatic production line capacity prediction device, the computer readable storage medium and the electronic equipment provided by the embodiments of the present application correspond to the automatic production line capacity prediction method, and the explanation, examples, beneficial effects and other parts of the related content can refer to the corresponding content in the automatic production line capacity prediction method, which will not be repeated here.

[0210] In summary, compared with the prior art, the present application has the following beneficial effects:

[0211] 1) The traditional capacity evaluation method is difficult to adapt to the characteristics of the surface mount production line dynamic and high-dimensional data. The present application is based on various historical characteristic data of the surface mount line equipment, extracts key features affecting the capacity through mutual information, constructs a PSO-ELM prediction model, realizes intelligent prediction of the production line capacity, and lays a foundation for fine and intelligent scheduling optimization and capacity planning.

[0212] 2) The existing data-driven production line capacity prediction model mainly considers the equipment failure time or downtime that occurs after the event, ignores the unnecessary disturbance or even shutdown of the dynamic disturbance event data to the normal operation of the production line, and leads to inaccurate prediction model accuracy, which is difficult to provide a scientific basis for the overall efficiency improvement of the production line equipment, fine control of the manufacturing process, and dynamic allocation of manufacturing resources. The present application fully considers the dynamic disturbance event data to correct the prediction model, effectively improves the accuracy of the prediction model.

[0213] It should be noted that, through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software plus necessary universal hardware platforms. Based on such an understanding, the above technical solutions can be embodied in the form of a software product in essence or in the form of a part of the prior art that makes a contribution. The computer software product can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments. In this article, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply that there is any such actual relationship or order between these entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the processes, methods, articles or devices that include a series of elements not only include those elements, but also include other elements not explicitly listed or inherent to such processes, methods, articles or devices. Without more limitations, the element defined by the statement "includes a" does not exclude the presence of other identical elements in the process, method, article or device that includes the element.

[0214] The above embodiments are only used to illustrate the technical solutions of the present application, and not to limit it; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements for some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting the capacity of an automated production line, characterized in that, The method comprises: obtaining a plurality of historical feature data and a plurality of current feature data of the manufacturing process of the automated production line, and preprocessing the same; obtaining mutual information values of each preprocessed historical feature data and processing time of an order, and screening a plurality of key feature data according to the mutual information values; constructing a sample data set for training a production capacity prediction model based on the plurality of key feature data, and training the production capacity prediction model based on an optimization algorithm to obtain a trained production capacity prediction model; wherein during the training, the following steps are further included: Based on various current characteristic data of the automated production line manufacturing process, it is determined whether a dynamic disturbance event has occurred during the production cycle. If no event has occurred, the original predicted value output by the capacity prediction model is used as the final predicted value. If an event has occurred, the original predicted value y before the dynamic disturbance event is obtained. pre (t) and actual value y rea (t), and calculate the first residual value er(t)=y rea (t)-y pre (t); Obtain the original predicted value y after a dynamic disturbance event occurs. pre (T) and actual value y rea (T), and calculate the second residual value er(T)=y rea (T)-y pre (T); Determine whether er(t) and er(T) simultaneously meet the tolerance requirements. If they do, then set y pre (t) or y pre Any one of (T) is used as the final predicted value; otherwise, the residuals er(T) corresponding to the n time points before the dynamic disturbance event are obtained. i ), i = 1, 2, ..., n; n is the number of types of key feature data to be screened; and is used as the input variable of the capacity prediction model to obtain the predicted compensation value y. err (T); and the predicted compensation value y err (T) and the original predicted value y pre Adding (T) together yields the corrected prediction value y. final (T) and used as the final predicted value; The key characteristic data to be predicted is input into the capacity prediction model to obtain an original prediction value corresponding to the key characteristic data to be predicted. If no dynamic disturbance event occurs in the production cycle or the tolerance requirement is met, the original prediction value output by the capacity prediction model is taken as the final prediction value. Otherwise, the corrected prediction value y final (T) is taken as the final prediction value.

2. The method of claim 1, wherein, The preprocessing includes missing value completion and data normalization.

3. The method of claim 1, wherein, The production capacity prediction model is a prediction model based on a limit learning machine, and the production capacity prediction model is optimized by using a particle swarm optimization algorithm; and the capacity prediction model is: wherein, x t an input vector representing a capacity prediction model; β i denote output weights, in total L; g() represents an activation function; ω i denote input weights, in total L; b i denotes the bias of the hidden layer node; y t representing x t corresponding output vector.

4. The method of claim 3, wherein, The objective function is taken as the fitness function of the particle swarm optimization algorithm; and the objective function f: min f = ‖βH - Y‖ 2 β = H + Y wherein β represents an output weight matrix, β = [β1, β2, …, βL] ; βi represents an output weight of the capacity prediction model, a total of L; b represents a bias of the hidden layer node, a total of L; i = 1, 2, …, L; and L represents the number of hidden layer nodes. i ,…β L ] ; β i i represents an output weight of the capacity prediction model, a total of L; b i i represents a bias of the hidden layer node, a total of L; i = 1, 2, …, L; and L represents the number of hidden layer nodes. x tn denotes the corresponding nth key feature data; H represents a matrix; H + is the Moore-Penrose generalized inverse of the matrix H; Y represents a true value.

5. The method of claim 4, wherein, The updating method of the speed and position in the particle swarm optimization algorithm is: wherein, is the motion speed of the i th particle in the k th and k+1 th iterations, respectively; ω is an inertia weight; c1 and c2 are learning factors; r1 and r2 are random numbers between 0 and 1; is the historical optimal position of the i th particle in the k th iteration; is the historical optimal position of the group in the k th iteration; is the position of the i th particle in the k th and k+1 th iterations, respectively.

6. The method of claim 5, wherein, The method for calculating the inertial weight ω is: where f i represents the objective function value of the i-th particle; f min denotes the minimum value in the history of the objective function values; f avg an average of historical values of the objective function; ω max denotes the maximum weight; ω min denotes the minimum weight.

7. The method of claim 1, wherein, During the training, the mean absolute percentage error, the mean absolute error, and the normalized mean square error are used as the evaluation criteria of the production capacity prediction model.

8. An apparatus for predicting the capacity of an automated production line, characterized in that, The device comprises: a data acquisition and preprocessing module configured to obtain a plurality of historical feature data and a plurality of current feature data of the manufacturing process of the automated production line, and to preprocess the same; a key feature data screening module configured to obtain mutual information values of each preprocessed historical feature data and processing time of an order, and to screen a plurality of key feature data according to the mutual information values; a production capacity prediction model training module configured to construct a sample data set for training a production capacity prediction model based on the plurality of key feature data, and to train the production capacity prediction model based on an optimization algorithm to obtain a trained production capacity prediction model; wherein during the training, the following steps are further included: Based on a plurality of current feature data of the automated production line manufacturing process, it is determined whether a dynamic disturbance event occurs within a production cycle; if not, the original predicted value output by the production capacity prediction model is taken as the final predicted value, and if so, the original predicted value y pre (t) before the dynamic disturbance event occurs is obtained rea (t) and a first residual value er(t) = y rea (t) - y pre (t) is calculated; the original predicted value y pre (T) after the dynamic disturbance event occurs is obtained rea (T) and a second residual value er(T) = y rea (T) - y pre (T) is calculated; it is determined whether er(t) and er(T) satisfy the tolerance requirement at the same time, if so, any one of y pre (t) or y pre (T) is taken as the final predicted value; otherwise, the residual er(T i ), i = 1, 2, …, n corresponding to the n time points before the dynamic disturbance event occurs are obtained; n is the number of types of key feature data for screening; and taken as the input variable of the production capacity prediction model, a predicted compensation value y err (T) is obtained; the predicted compensation value y err (T) is added to the original predicted value y pre (T) to obtain a corrected predicted value y final (T) and taken as the final predicted value. The prediction result output module is configured to input the to-be-predicted key feature data into the production capacity prediction model to obtain an original prediction value corresponding to the to-be-predicted key feature data, and if no dynamic disturbance event occurs within a production cycle or a tolerance requirement is met, the original prediction value output by the production capacity prediction model is taken as a final prediction value, otherwise, a corrected prediction value y final (T) as the final prediction value.

9. A computer-readable storage medium, characterized in that, The computer program for predicting the production capacity of the automated production line, wherein the computer program enables the computer to execute the automated production line production capacity prediction method according to any one of claims 1-7.

10. An electronic device comprising: one or more processors; memory; and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, characterized in that the programs include a program for executing the automated production line production capacity prediction method according to any one of claims 1-7.

Images