Standardized management method in intelligent production process of mechanical and electrical products

By dividing the electromechanical product manufacturing process into sub-steps and utilizing a deep learning network model, the problems of low production efficiency and inaccurate data under traditional manual monitoring have been solved. This has enabled real-time monitoring and intelligent adjustment of the electromechanical product manufacturing process, thereby improving production efficiency and product quality.

CN118897520APending Publication Date: 2024-11-05CHINA NAT INST OF STANDARDIZATION
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410926893.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-11
Publication Date
2024-11-05

AI Technical Summary

Technical Problem

In the traditional electromechanical product manufacturing process, standardized management relies on manual monitoring, which leads to low production efficiency, inaccurate data, difficulty in responding to production changes in real time, inability to effectively predict future production conditions, and inability to fully utilize Internet of Things and big data technologies.

Method used

The assembly process of electromechanical products is divided into multiple sub-steps. Production data is analyzed and predicted in real time through deep learning network models, and control parameters are automatically adjusted. Data processing and prediction are performed using LSTM units to ensure the stability and consistency of the production process.

Benefits of technology

It enables real-time monitoring and intelligent adjustment of the production process, improves production efficiency and product quality, reduces manual intervention, provides a scientific basis for production adjustment, and enhances the level of intelligent manufacturing.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118897520A_ABST
    Figure CN118897520A_ABST
Patent Text Reader

Abstract

The invention discloses a standardized management method in an electromechanical product intelligent production process, and the method comprises the steps: dividing an electromechanical product assembly process into n steps, determining the optimal and acceptable control parameter range of each step, carrying out the normalization of the optimal and acceptable control parameter range into standard data, collecting the control parameters of each step in real time, forming time window data, carrying out the normalization of the time window data, and constructing a deep learning model. Actual data is trained to be optimal, the model predicts data of the next time window, the change rate is compared with current monitoring data, corresponding step parameters are adjusted if the change rate exceeds a threshold value, and then the predicted data is comprehensively adjusted to be optimal parameters if the similarity of the predicted data and standard data is lower than the threshold value; according to the method, the stability and consistency of the production process can be ensured, the production efficiency, the product quality and the intelligent level are remarkably improved, and powerful support is provided for intelligent manufacturing of mechanical and electrical products.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of standardization management, and in particular relates to a standardization management method in the intelligent production process of electromechanical products. Background Art

[0002] In the current production process of electromechanical products, the traditional standardized management method relies on manual monitoring and adjustment. This method has many significant disadvantages. Manual monitoring requires a lot of time and human resources, and it is difficult to respond to changes in the production process in real time, resulting in low production efficiency. Human factors may cause inaccurate monitoring data, which in turn affects the adjustment of production control parameters and increases product rejection rates. Traditional methods can usually only monitor the current production status, but cannot effectively predict future production conditions, making it difficult to take preventive measures in advance.

[0003] Under the trend of intelligent manufacturing, as the scale of production expands, traditional methods are unable to cope with large amounts of data, and traditional management methods cannot fully utilize advanced technologies such as the Internet of Things, big data, and deep learning to achieve comprehensive intelligence in the production process. How to solve the standardized management of the intelligent production process of electromechanical products and improve the stability and consistency of the production process is a problem that needs to be solved. Summary of the invention

[0004] The purpose of the present invention is to provide a standardized management method in the intelligent production process of electromechanical products, which divides the standardized production process of electromechanical product assembly into multiple sub-steps, and performs real-time adjustment of production optimization control parameters based on the monitoring data and prediction data of the sub-steps. The specific technical scheme is as follows:

[0005] The standardized management method includes the following steps:

[0006] Step S1, divide the standardized production process of electromechanical product assembly into n sub-steps, and determine the optimal control parameters and the range of acceptable control parameters for each sub-step, normalize the optimal control parameters to data of 0 to 1, and obtain standard data for production control.

[0007] Step S2, continuously collect the actual production control parameters of each sub-step on the product line in multiple time windows, form an actual parameter set based on the time window, and normalize the actual parameter set into actual data based on the time window sequence.

[0008] Step S3, constructing a deep learning network model, and using the actual data as training data to train the deep learning network model until a trained network model is obtained.

[0009] Step S4, based on the trained network model, using the monitoring data corresponding to the monitoring parameters of the current time window as input, and obtaining the predicted data of the next time window through model prediction.

[0010] Step S5, compare the predicted data of the next time window of each sub-step with the monitoring data corresponding to the current time window, calculate the data change rate, and if the data change rate of a sub-step exceeds the preset threshold, adjust the production control parameters of the sub-step to the control parameters monitored in the current time window.

[0011] Step S6, comparing the monitoring data formed in the next time window with the standard data for similarity, if the similarity is less than a preset similarity threshold, adjusting the production control parameters of all sub-steps of the electromechanical product assembly to the optimal control parameters.

[0012] Further, the standardized production process of electromechanical product assembly is divided into n sub-steps, wherein the sub-steps at least include: an assembly step, a debugging step, and an inspection step;

[0013] The optimal control parameters are normalized to data of length 0 to 1, and the normalization method is:

[0014]

[0015] Among them, a i is the normalized data of the ith sub-step, is the optimal control parameter of the ith sub-step, min(m i ) and max(m i ) are the lower and upper limits of the control parameters corresponding to the ith sub-step when it does not affect the product quality;

[0016] The standard data is expressed as: A0=(a1, a2, ..., a i , ...a n ), where 0 < a i <1,n≥3.

[0017] Furthermore, the actual production control parameters of each sub-step on the product line in multiple time windows are continuously collected to form an actual parameter set B based on the time window = {B1, B2, ..., B j , ..., B s}; where s is the number of time windows, B j represents the production control parameter of the jth time window, B j =(n j1 , n j2 , ..., n ji ,...n jn ), n jiis the production control parameter of the i-th sub-step in the j-th time window, and the production control parameter is the average value of the production control parameters collected by the sensor in each time window;

[0018] Then the actual data b of the jth time window j =(b j1 , b j2 , ..., b ji , ...b jn ), b ji The actual data of the i-th sub-step in the j-th time window is:

[0019] Furthermore, the deep learning network model includes an input layer, a recurrent layer and an output layer;

[0020] Take the actual data as the input of the input layer. At time t, the input of the input layer is recorded as x t ,The recurrent layer uses a long short-term memory network composed of LSTM units, and adds a forget gate, an input gate, and an output gate to each LSTM unit;

[0021] The forget gate is used to control the memory unit c at time t-1 t-1 For the memory unit c at time t t The control variable of the forget gate is f(t), f(t) = σ(W fh h t-1 +W fx x t +b f ), where h t-1 is the output of the recurrent layer at time t-1, W fh is the weight of the recurrent layer output in the forget gate, x t is the input layer at time t, W fx is the weight of the input layer in the forget gate, b f is the bias term in the forget gate, σ is the activation function Sigmoid;

[0022] The input gate is used to control the LSTM unit input x at time t t and the acceptance degree i(t) of the recurrent layer output ht-1 at time t-1, i(t) = σ(W ih h t-1 +W ix x t +b i ), W ih is the weight of the recurrent layer output in the input gate, W ix is the weight of the input layer in the input gate, b i is the bias term in the input gate;

[0023] Through the Tanh function on the input x t The nonlinear transformation is performed with the output ht-1 at time t-1, and the bias output is obtained through the bias output function d(t), d(t) = Tanh(W dh h t-1 +W dx x t +b d ), where W dh is the weight of the recurrent layer output in the bias output function, W dx is the weight of the input layer in the bias output function, b d is the bias term in the bias output function;

[0024] If the LSTM unit accepts the output bias d(t), then the memory unit c of the LSTM at time t t =f(t)c t-1 +i(t)d(t), if the LSTM unit accepts the output bias d(t), then the memory unit c of the LSTM at time t t =f(t)c t-1 ;

[0025] The output gate is used to control the LSTM unit to the memory cell c at time t t The acceptance of the control variable o(t) = σ(W oh h t-1 +Woxxt+bo, where Woh is the weight of the output of the loop layer in the output gate, Wox is the weight of the input of the input layer in the output gate, and b o is the bias term in the output gate;

[0026] Then the output h of the LSTM unit at time t is t =o(t)*Tanh(c t ), where * represents the Hadamard matrix product operation.

[0027] Furthermore, in the forget gate, when f(t)=1, the forget gate is opened to accept the memory unit c at the previous moment. t-1 When f(t) = 0, the forget gate is closed and the memory unit c of the previous moment is not accepted. t-1 any information;

[0028] In the input gate, when i(t)=1, the LSTM unit accepts the output bias d(t), and when i(t)=0, the input gate is closed and does not accept the output bias d(t);

[0029] In the output gate, when o(t)=1, the output gate accepts the memory cell c t When o(t) = 0, the output gate is closed and does not accept the memory unit c.t any information.

[0030] Furthermore, based on the trained network model, the monitoring data b collected in the current time window is input x =(b x1 , b x2 , ..., b xi , ...b xn ), then we get the predicted data b′x for the next time window +1 =(b′x +1 , 1, b′x +1 , 2, ..., b′x +1 ,i,...b′x +1 , n);

[0031] Then the data change rate of the predicted data of the next time window of the i-th sub-step and the monitoring data corresponding to the current time window is The preset threshold value of the change rate is 0.2% to 0.5%.

[0032] Furthermore, the monitoring data formed in the next time window is compared with the standard data for similarity, and the predicted data b′ of the next time window is x+1 =(b′ x+1,1 , b′ x+1,2 , ..., b′ x+1,i , ...b′ x+1,n ), standard data is: A0=(a1,a2,...,a i , ...a n ), then the similarity calculation formula is:

[0033]

[0034] S x+1 is the similarity between the monitoring data and the standard data formed in the next time window. The preset similarity threshold S0 is not less than 0.95. x+1 <S0, then adjust the production control parameters of all sub-steps of electromechanical product assembly to the optimal control parameters.

[0035] Furthermore, after adjusting the production control parameters of all sub-steps of the electromechanical product assembly to the optimal control parameters, steps S2 to S6 are repeated to execute a new round of standardized management in the intelligent production process of electromechanical products.

[0036] Furthermore, based on a preset time period or when the standardized production process of electromechanical product assembly changes, steps S1 to S6 are re-executed, and the preset time period is 30 to 90 days.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention uses a deep learning network model to perform real-time analysis and prediction of production data, can automatically adjust production control parameters, and reduce manual intervention. At the same time, the deep learning network model can provide more accurate production data, predict possible problems in advance, provide a scientific basis for production adjustments, and avoid production delays and losses. The method of the present invention can process large-scale production data, realize real-time monitoring and intelligent adjustment of the production process, ensure the stability and consistency of the production process, significantly improve production efficiency, product quality and intelligence level, and provide strong support for the intelligent manufacturing of electromechanical products. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flow chart of a standardized management method in the intelligent production process of electromechanical products.

[0040] Figure 2 Schematic diagram of the deep learning network model structure of the present invention.

[0041] Figure 3 Schematic diagram of the LSTM unit structure of the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are 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 are within the scope of protection of the present invention.

[0043] like Figure 1 FIG. 1 is a flow chart of a standardized management method for intelligent production of electromechanical products according to the present invention, and the method comprises the following steps:

[0044] Step S1, divide the standardized production process of electromechanical product assembly into n sub-steps, and determine the optimal control parameters and the range of acceptable control parameters for each sub-step, normalize the optimal control parameters to data of 0 to 1, and obtain standard data for production control.

[0045] It should be noted that the standardized production process of electromechanical product assembly can be divided into n sub-steps and preliminarily classified according to the role of each production link in the realization of the function of the electromechanical product; for example, the assembly of components directly related to the core function of the product can be classified as an "assembly step"; and activities related to product performance debugging are classified as "debugging steps", and the subsequent inspection process is divided into: "inspection steps"; the sub-steps include at least: assembly steps, debugging steps and inspection steps.

[0046] Exemplarily, based on the above classification method, the standardized production process for assembling a certain electromechanical product (a certain model of electric motor) can be divided into the following five specific sub-steps: Sub-step 1 - Parts preparation, pick up all the parts required for the motor from the warehouse, including the stator, rotor, bearings, end covers, fans, junction boxes, etc., perform visual inspection on the parts to ensure that there are no defects such as damage, rust, and deformation, clean, degrease, and rust-remove the parts that need pretreatment to ensure that the surface of the parts is clean and free of impurities, measure the dimensions and check the accuracy of key parts to ensure that they meet the design requirements; Sub-step 2 - Preliminary assembly, fix the stator on the assembly table, install the bearings and end covers, insert the rotor into the stator, ensure that there is an appropriate gap with the stator, install the fan and junction box, connect the internal lines, and use special tools to perform preliminary tightening of the assembled parts to ensure that the components are firmly connected; Sub-step 3 - Function debugging, connect the power supply, start the motor, observe its operation, check the motor's speed, torque, vibration and other performance indicators, use test instruments to test the motor's electrical performance, including voltage, current, power factor, etc., adjust the motor's parameters such as voltage, current, frequency, etc. according to the test results to achieve the best working state, conduct long-term operation tests, and observe the stability and durability of the motor; Sub-step 4--quality inspection, conduct a comprehensive inspection of the assembled motor, including appearance quality, assembly quality, functional performance, etc., use automatic light sensing measurement tools to review the key dimensions of the motor, conduct necessary insulation resistance tests and withstand voltage tests, and mark and handle unqualified items; Sub-step 5--final packaging, clean the qualified motor, remove surface oil and impurities, and mark the motor model, specifications, production date, manufacturer and other information on the package.

[0047] After determining these sub-steps, it is necessary to determine the optimal control parameters (such as assembly accuracy, debugging time, etc.) and the acceptable range of control parameters for each sub-step, and normalize these parameters to data between 0 and 1 for subsequent data processing and comparison.

[0048] The optimal control parameters are normalized to data of length 0 to 1, and the normalization method is:

[0049]

[0050] Among them, a i is the normalized data of the ith sub-step, is the optimal control parameter of the ith sub-step, min(m i ) and max(m i ) are the lower and upper limits of the control parameters corresponding to the ith sub-step when it does not affect the product quality;

[0051] The standard data is expressed as: A0=(a1, a2, ..., a i , ...a n ), where 0 < a i <1,n≥3.

[0052] Assume that we have determined the optimal control parameters for each sub-step and normalized them. The normalized standard data: A0 = (a1, a2, a3, a4, a5) = (0.2, 0.5, 0.7, 0.9, 0.6).

[0053] Step S2, continuously collect the actual production control parameters of each sub-step on the product line in multiple time windows, form an actual parameter set based on the time window, and normalize the actual parameter set into actual data based on the time window sequence.

[0054] It should be noted that the actual production control parameters of each sub-step on the product line in multiple time windows are continuously collected to form an actual parameter set B based on the time window = {B1, B2, ..., B j , ..., B s}; where s is the number of time windows, B j represents the production control parameter of the jth time window, B j =(n j1 , n j2 , ..., n ji ,...n jn ), n ji is the production control parameter of the ith sub-step in the jth time window, and the production control parameter is the average value of the production control parameters collected by the sensor in each time window.

[0055] Then the actual data b of the jth time window j =(b j1 , b j2 , ..., b ji , ...b jn ), b ji The actual data of the i-th sub-step in the j-th time window is:

[0056] For example, suppose we continuously collect actual production control parameters of three time windows (j=1, 2, 3) and perform normalization processing, and the actual parameter set is: B={B1, B2, B3};

[0057] Among them, B1 = (0.21, 0.48, 0.69, 0.88, 0.59), B2 = (0.22, 0.51, 0.71, 0.91, 0.61), and B3 = (0.20, 0.49, 0.70, 0.89, 0.60).

[0058] Step S3, constructing a deep learning network model, and using the actual data as training data to train the deep learning network model until a trained network model is obtained.

[0059] like Figure 2 As shown, the deep learning network model includes an input layer, a loop layer and an output layer; the actual data (the monitoring data corresponding to the monitoring parameters of the current time window) is used as the input of the input layer. At time t, the input of the input layer is recorded as x t The recurrent layer uses a long short-term memory network composed of LSTM units, and adds a forget gate, an input gate, and an output gate to each LSTM unit; the output layer is used to output the predicted data of the next time window.

[0060] like Figure 3 As shown, the forget gate is used to control the memory unit c at time t-1. t-1 For the memory unit c at time t t The control variable of the forget gate is f(t), f(t) = σ(W fh h t-1 +W fx x t +b f ), where h t-1 is the output of the recurrent layer at time t-1, W fh is the weight of the recurrent layer output in the forget gate, x t is the input layer at time t, W fx is the weight of the input layer in the forget gate, b f is the bias term in the forget gate, and σ is the activation function Sigmoid.

[0061] The input gate is used to control the input x of the LSTM unit at time t t and the recurrent layer output h at time t-1 t-1 The acceptance degree i(t), i(t) = σ(W ih h t-1 +W ix x t +b i ), W ih is the weight of the recurrent layer output in the input gate, W ix is the weight of the input layer in the input gate, b i is the bias term in the input gate.

[0062] Through the Tanh function on the input x t And the output h at time t-1 t-1 Perform nonlinear transformation and obtain bias output through bias output function d(t), d(t) = Tanh(W dh h t-1 +W dx x t +b d ), where W dh is the weight of the recurrent layer output in the bias output function, W dx is the weight of the input layer in the bias output function, b d is the bias term in the bias output function.

[0063] If the LSTM unit accepts the output bias d(t), then the memory unit c of the LSTM at time t t =f(t)c t-1 +i(t)d(t), if the LSTM unit accepts the output bias d(t), then the memory unit c of the LSTM at time t t =f(t)c t-1 .

[0064] The output gate is used to control the LSTM unit to the memory cell c at time t t The acceptance of the control variable o(t) = σ(W oh h t-1 +W ox x t +b o ), where W oh is the weight of the output of the recurrent layer in the output gate, W ox is the weight of the input layer in the output gate, b o is the bias term in the output gate.

[0065] Then the output h of the LSTM unit at time t is t =o(t)*Tanh(c t ), where * represents the Hadamard matrix product operation.

[0066] In the forget gate, when f(t)=1, the forget gate is opened to accept the memory unit c at the previous moment. t-1 When f(t) = 0, the forget gate is closed and the memory unit c of the previous moment is not accepted. t-1 any information.

[0067] In the input gate, when i(t) = 1, the LSTM unit accepts the output bias d(t), and when i(t) = 0, the input gate is closed and does not accept the output bias d(t); in the output gate, when o(t) = 1, the output gate accepts the memory unit c t When o(t) = 0, the output gate is closed and does not accept the memory unit c. t any information.

[0068] Step S4, based on the trained network model, using the monitoring data corresponding to the monitoring parameters of the current time window as input, and obtaining the predicted data of the next time window through model prediction.

[0069] Based on the trained network model, input the monitoring data b collected in the current time window x =(b x1 , b x2 , ..., b xi , ...b xn ), then the predicted data b′ of the next time window is obtained x+1 =(b′ x+1,1 , b′ x+1,2 , ..., b′ x+1,i , ...b′ x+1,n ).

[0070] For example, assuming that the current time window is the third time window, we input the normalized monitoring data b3 of the time window for prediction: b3 = (0.20, 0.49, 0.70, 0.89, 0.60); through the deep learning model prediction, the normalized production control parameters of the next time window are: b′4 = (b′ 41 , b′ 42 , b′ 43 , b′ 44 , b′ 45 )=(0.21, 0.50, 0.71, 0.90, 0.62).

[0071] Step S5, compare the predicted data of the next time window of each sub-step with the monitoring data corresponding to the current time window, calculate the data change rate, and if the data change rate of a sub-step exceeds the preset threshold, adjust the production control parameters of the sub-step to the control parameters monitored in the current time window.

[0072] Then the data change rate of the predicted data of the next time window of the i-th sub-step and the monitoring data corresponding to the current time window is The preset threshold value of the change rate is 0.2% to 0.5%.

[0073] For example, we calculate the data change rate ρ between the predicted data of the next time window of the i-th sub-step and the monitoring data corresponding to the current time window bi :

[0074] For substep 1: b1 = 0.5%, for sub-step 2: ρ b2 =0.2%, and so on to calculate the change rate of other sub-steps; assuming that the preset threshold is 0.3%, the change rate of sub-step 1 exceeds the threshold, and its production control parameters need to be adjusted.

[0075] Step S6, comparing the monitoring data formed in the next time window with the standard data for similarity, if the similarity is less than a preset similarity threshold, adjusting the production control parameters of all sub-steps of the electromechanical product assembly to the optimal control parameters.

[0076] The monitoring data formed in the next time window is compared with the standard data for similarity, and the predicted data b′ in the next time window is x+1 =(b′ x+1,1 , b′ x+1,2 , ..., b′ x+1,i, ...b′ x+1,n ), standard data is: A0=(a1,a2,...,a i , ...a n ), then the similarity calculation formula is:

[0077]

[0078] S x+1 is the similarity between the monitoring data and the standard data formed in the next time window. The preset similarity threshold S0 is not less than 0.95. x+1 <S0, then adjust the production control parameters of all sub-steps of electromechanical product assembly to the optimal control parameters.

[0079] It should be noted that step S5 and step S6 are parallel steps and can be interchanged. Generally, when S x+1 <S0, then in step S5 there must be a sub-step whose data change rate exceeds the preset threshold.

[0080] After the production control parameters of all sub-steps of the electromechanical product assembly are adjusted to the optimal control parameters, steps S2 to S6 are repeated to execute a new round of standardized management in the intelligent production process of electromechanical products.

[0081] Based on a preset time period or when the standardized production process of electromechanical product assembly changes, steps S1 to S6 are re-executed, and the preset time period is 30 to 90 days.

[0082] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A standardized management method in the intelligent production process of electromechanical products, characterized in that: The method comprises the following steps: Step S1, dividing the standardized production process of electromechanical product assembly into n sub-steps, and determining the optimal control parameters and the range of acceptable control parameters for each sub-step, normalizing the optimal control parameters to data between 0 and 1, and obtaining standard data for production control; Step S2, continuously collecting actual production control parameters of each sub-step on the product line within multiple time windows to form an actual parameter set based on the time window, and normalizing the actual parameter set into actual data based on the time window sequence; Step S3, constructing a deep learning network model, and using the actual data as training data to train the deep learning network model until a trained network model is obtained; Step S4, based on the trained network model, using the monitoring data corresponding to the monitoring parameters of the current time window as input, and obtaining the predicted data of the next time window through model prediction; Step S5, compare the predicted data of the next time window of each sub-step with the monitoring data corresponding to the current time window, calculate the data change rate, and if the data change rate of a sub-step exceeds a preset threshold, adjust the production control parameter of the sub-step to the control parameter monitored in the current time window; Step S6, comparing the monitoring data formed in the next time window with the standard data for similarity, if the similarity is less than a preset similarity threshold, adjusting the production control parameters of all sub-steps of the electromechanical product assembly to the optimal control parameters.

2. A standardized management method in the intelligent production process of electromechanical products according to claim 1, characterized in that: Divide the standardized production process of electromechanical product assembly into n sub-steps, wherein the sub-steps at least include: an assembly step, a debugging step, and an inspection step; The optimal control parameters are normalized to data of length 0 to 1, and the normalization method is: Among them, a i is the normalized data of the ith sub-step, is the optimal control parameter of the ith sub-step, min(m i ) and max(m i ) are the lower and upper limits of the control parameters corresponding to the ith sub-step when it does not affect the product quality; The standard data is expressed as: A0=(a1, a2, ..., a i , ...a n ), where 0 < a i <1,n≥3.

3. The standardized management method in the intelligent production process of electromechanical products according to claim 2 is characterized in that: Continuously collect the actual production control parameters of each sub-step on the product line in multiple time windows to form an actual parameter set B based on the time window = {B1, B2, ..., B j , ..., B s }; where s is the number of time windows, B j represents the production control parameter of the jth time window, B j =(n j1 , n j2 , ..., n ji ,...n jn ), n ji is the production control parameter of the i-th sub-step in the j-th time window, and the production control parameter is the average value of the production control parameters collected by the sensor in each time window; Then the actual data b of the jth time window j =(b j1 , b j2 , ..., b ji , ...b jn ), b ji The actual data of the i-th sub-step in the j-th time window is:

4. A standardized management method in the intelligent production process of electromechanical products according to claim 3, characterized in that: The deep learning network model includes an input layer, a circulation layer and an output layer; Take the actual data as the input of the input layer. At time t, the input of the input layer is recorded as x t ,The recurrent layer uses a long short-term memory network composed of LSTM units, and adds a forget gate, an input gate, and an output gate to each LSTM unit; The forget gate is used to control the memory unit c at time t-1 t-1 For the memory unit c at time t t The control variable of the forget gate is f(t), f(t) = σ(W fh h t-1 +W fx x t +b f ), where h t-1 is the output of the recurrent layer at time t-1, W fh is the weight of the recurrent layer output in the forget gate, x t is the input layer at time t, W fx is the weight of the input layer in the forget gate, b f is the bias term in the forget gate, σ is the activation function Sigmoid° The input gate is used to control the LSTM unit input x at time t t and the acceptance degree i(t) of the recurrent layer output ht-1 at time t-1, i(t) = σ(W ih h t-1 +W ix x t +b i ), W ih is the weight of the recurrent layer output in the input gate, W ix is the weight of the input layer in the input gate, b i is the bias term in the input gate; Through the Tanh function on the input x t The nonlinear transformation is performed with the output ht-1 at time t-1, and the bias output is obtained through the bias output function d(t), d(t) = Tanh(W dh h t-1 +W dx x t +b d ), where W dh is the weight of the recurrent layer output in the bias output function, W dx is the weight of the input layer in the bias output function, b d is the bias term in the bias output function; If the LSTM unit accepts the output bias d(t), then the memory unit c of the LSTM at time t t =f(t)c t-1 +i(t)d(t), if the LSTM unit accepts the output bias d(t), then the memory unit c of the LSTM at time t t =f(t)c t-1 ; The output gate is used to control the LSTM unit to the memory cell c at time t t The acceptance of the control variable o(t) = σ(W oh h t-1 +Woxxt+bo, where Woh is the weight of the output of the loop layer in the output gate, Wox is the weight of the input of the input layer in the output gate, and b o is the bias term in the output gate; Then the output h of the LSTM unit at time t is t =o(t)*Tanh(c t ), where * represents the Hadamard matrix product operation.

5. A standardized management method in the intelligent production process of electromechanical products according to claim 4, characterized in that: In the forget gate, when f(t)=1, the forget gate is opened to accept the memory unit c at the previous moment. t-1 When f(t) = 0, the forget gate is closed and the memory unit c of the previous moment is not accepted. t-1 any information; In the input gate, when i(t)=1, the LSTM unit accepts the output bias d(t), and when i(t)=0, the input gate is closed and does not accept the output bias d(t); In the output gate, when o(t)=1, the output gate accepts the memory cell c t When o(t) = 0, the output gate is closed and does not accept the memory unit c. t any information.

6. A standardized management method in the intelligent production process of electromechanical products according to claim 5, characterized in that: Based on the trained network model, input the monitoring data b collected in the current time window x =(b x1 , b x2 , ..., b xi ,...bx n ), then the predicted data b′ of the next time window is obtained x+1 =(b′ x+1 , 1, b′ x+1 , 2, ..., b′ x+1,i , ...b′ x+1,n ); Then the data change rate of the predicted data of the next time window of the i-th sub-step and the monitoring data corresponding to the current time window is The preset threshold value of the change rate is 0.2% to 0.5%.

7. The standardized management method in the intelligent production process of electromechanical products according to claim 5 is characterized in that: The monitoring data formed in the next time window is compared with the standard data for similarity, and the predicted data b′ in the next time window is x+1 =(b′ x+1 , 1, b′ x+1,2 , ..., b′ x+1,i , ...b′ x+1,n ), standard data is: A0=(a1,a2,...,a i , ...a n ), then the similarity calculation formula is: S x+1 is the similarity between the monitoring data and the standard data formed in the next time window. The preset similarity threshold S0 is not less than 0.

95. x+1 <S0, then adjust the production control parameters of all sub-steps of electromechanical product assembly to the optimal control parameters.

8. A standardized management method in the intelligent production process of electromechanical products according to claim 7, characterized in that: After the production control parameters of all sub-steps of the electromechanical product assembly are adjusted to the optimal control parameters, steps S2 to S6 are repeated to execute a new round of standardized management in the intelligent production process of electromechanical products.

9. A standardized management method in the intelligent production process of electromechanical products according to claim 8, characterized in that: Based on a preset time period or when the standardized production process of electromechanical product assembly changes, steps S1 to S6 are re-executed, and the preset time period is 30 to 90 days.