Winding machine capacity alarm efficiency evaluation method

Abstract

The invention discloses a method for evaluating productivity alarm efficiency of a winding machine, which comprises the following steps: an alarm screening module is used for determining a main influence factor of a single coil NG, acquiring corresponding influence factor data, cleaning the acquired data, taking the influence factor as an alarm factor and the productivity as an output prediction quantity, and determining the productivity of the winding machine; establishing a data wide table of an input-output relationship; screening features with high importance for feature factors in the data wide table based on the importance of a lightgbm tree model, and giving a factor importance sequence by using MSE as the basis of node splitting; and the DEA efficiency evaluation module obtains the comprehensive efficiency through CCR calculation and obtains the technical efficiency through BCC calculation, and the scale efficiency = the comprehensive efficiency / the technical efficiency. The invention provides a winding machine capacity alarm efficiency evaluation method, which is used for positioning a top factor influencing the capacity and the yield, assisting a service output maintenance list, and assisting the service output maintenance list by utilizing a DEA efficiency evaluation method.

Description

Technical Field

[0001] The present invention relates to the technical field of battery production efficiency evaluation, and in particular to a method for evaluating the production capacity alarm efficiency of a winding machine. Background Art

[0002] In the production and processing of batteries, the multi-layer materials of the core need to be wound and formed. During the operation of the winding machine, the alarm abnormality triggers a single roll NG. The common single roll NG alarms include: abnormal cutting position, tab error, edge alarm, insertion length alarm, etc. The scrapping of a single roll NG will inevitably lead to a decrease in production capacity. In addition, the winding process is a key process that affects production capacity. Currently, there are multiple winding machines on each line, and the number of alarms per day is as high as 1.4w / day. The single roll NG alarm of the winding process is the main factor affecting production capacity and yield.

[0003] The existing winding capacity alarm has the following problems: the top alarms that affect the capacity and yield of multiple machines are not clearly positioned, there is a lack of quantitative means, and there is a lack of quantitative means for the improvement analysis of the differences among multiple machines. Therefore, it is necessary to be able to evaluate the efficiency of the capacity alarm and explore how to improve it. DEA is an efficiency evaluation method that calculates multiple inputs and multiple outputs. It measures the relative efficiency of the decision-making unit (DMU) based on linear programming by constructing the production frontier and gives the improvement adjustment amount. The decision-making unit refers to the individual whose efficiency needs to be evaluated. However, there is no relevant existing technology on how to quantify the machine differences for the winding machine, such as the winding machine, coating roll, and slurry pot number. Summary of the invention

[0004] The purpose of the present invention is to provide a method for evaluating the efficiency of the production capacity alarm of a winding machine, using statistical means to analyze the relationship between the number of alarms in each process step, the shutdown time, and the production capacity and yield, locate the top factors affecting the production capacity and yield, assist the business in outputting a maintenance list, and use the DEA efficiency evaluation method to assist the business in outputting a maintenance list.

[0005] The technical solution adopted by a winding machine production capacity alarm efficiency evaluation method disclosed in the present invention is:

[0006] A method for evaluating the efficiency of a winding machine production capacity alarm, comprising:

[0007] The alarm screening module determines the main influencing factors of single-volume NG, obtains the corresponding influencing factor data, cleans the obtained data, and uses the influencing factors as alarm factors and the production capacity as the output prediction quantity, thereby establishing a data wide table with input-output relationship; based on the importance of the lightgbm tree model, the characteristic factors in the data wide table are screened for features with high importance, and MSE is used as the basis for node splitting. In order to evaluate the quantitative relationship between the importance of the screened factors and the production capacity, the input-output relationship is fitted based on Light-GBM, and then the factor importance ranking is given under the premise of ensuring the fitting accuracy;

[0008] DEA efficiency evaluation module, based on influencing factors and wide table data as the input of the efficiency evaluation calculation module, CCR calculation obtains the comprehensive efficiency, BCC calculation obtains the technical efficiency, and scale efficiency = comprehensive efficiency / technical efficiency, where technical efficiency: reflects the efficiency loss caused by input factors under the current input status; scale efficiency: reflects the efficiency loss caused by output factors under the current output status.

[0009] As a preferred scheme, the influencing factors include: abnormal positive electrode cutting position, positive electrode ear error, positive electrode single-pole winding, abnormal negative electrode cutting position, negative electrode material cutting, negative electrode ear error, negative electrode edge alarm, diaphragm hot hole error, positive electrode insertion length alarm, negative electrode insertion length alarm, positive electrode covering tape 2 error, positive electrode covering tape 2 contact check, poor CCD detection, abnormal rear winding length, and poor collection machine.

[0010] As a preferred solution, the data is cleaned mainly to delete records of downtime greater than 3 hours during the maintenance period and abnormal production capacity records.

[0011] As a preferred solution, the calculation formula of the MSE is:

[0012]

[0013] Among them, N is the number of samples of the current node, y i is the current node value, node prediction value MSE=(y i -mean) 2 / N.

[0014] As a preferred solution, the screened factors are mainly selected, and the top 6 factors after MSE calculation are output as important factors.

[0015] As a preferred solution, the Light-GBM obtains a capacity prediction model based on characteristic input factors by constructing the relationship between characteristics and capacity compensation capacity. It uses grid adjustment to adjust the key parameters affecting the model and obtains the adjustment parameters based on the MSE indicator.

[0016] As a preferred solution, the BCC model is as follows:

[0017]

[0018] The CCR model is similar to the BCC model, except that there are fewer constraints.

[0019] Where: θ is the efficiency coefficient, λ r is the slack adjustment amount and weight coefficient of the i-th input and j-th output, X i , Y j is the i-th input and j-th output variable.

[0020] As a preferred solution, the technical efficiency and scale efficiency have the following corresponding meanings:

[0021] When technical efficiency = 1, it means that the decision-making unit is in the optimal scale state;

[0022] When technical efficiency is <1, it means that the decision-making unit has unreasonable investment;

[0023] When scale efficiency = 1, it means that the decision-making unit is in the optimal scale state;

[0024] When the scale efficiency is <1, it means that the decision-making unit has unreasonable output.

[0025] The beneficial effects of a winding machine capacity alarm efficiency evaluation method disclosed in the present invention are: using statistical means to analyze the relationship between the number of alarms in each process step, the downtime and the capacity and yield, locating the top factors affecting the capacity and yield, assisting the business in outputting a maintenance list, and using the DEA efficiency evaluation method to assist the business in outputting a maintenance list. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a flow chart of a winding machine production capacity alarm efficiency evaluation method of the present invention. DETAILED DESCRIPTION

[0027] The present invention will be further described and illustrated below in conjunction with specific embodiments and accompanying drawings:

[0028] Please refer to Figure 1 , a winding machine production capacity alarm efficiency evaluation method, comprising:

[0029] The alarm screening module determines the main influencing factors of single-volume NG and obtains the corresponding influencing factor data.

[0030] In this embodiment, the current maintenance cycle is 15 days, there are 10 winding machines in the line, each machine produces in 2 shifts every day, and the minimum data granularity is the shift level. The specific characteristic factors are as follows:

[0031] Table 1: Characteristic factor table

[0032] Serial number Winding single roll NG influence factor 1 Abnormal cutting position of positive electrode 2 Positive tab error 3 Positive pole winding 4 Negative electrode cutting position 5 Negative electrode material cutting 6 Negative tab error 7 Negative edge alarm 8 Diaphragm hot hole error 9 Positive electrode insertion length alarm 10 Negative electrode insertion length alarm 11 Positive electrode covering tape 2 error 12 Positive electrode covering tape 2 contact inspection 13 CCD detection is poor 14 Abnormal length of rear winding (foil periphery) 15 Defective collection machine

[0033] The acquired data is cleaned and processed, and the influencing factors are used as alarm factors, and the production capacity is used as the output prediction quantity, so as to establish a data wide table of input-output relationship. The specific data wide table is as follows;

[0034]

[0035]

[0036] The feature factors in the data wide table are selected based on the importance of the lightgbm tree model, and MSE is used as the basis for node splitting.

[0037] The calculation formula of MSE is:

[0038]

[0039] Among them, N is the number of samples of the current node, y i is the current node value, node prediction value MSE=(y i -mean) 2 / N.

[0040] In order to evaluate the quantitative relationship between the importance of the screened factors and production capacity, the input-output relationship was fitted based on Light-GBM, and then the factor importance ranking was given while ensuring the fitting accuracy.

[0041] The top 6 factors after MSE calculation are output as important factors. After screening, the characteristic factors are mainly ranked as follows:

[0042] feature_names impt_percent Negative electrode material defective tape 0.24 Positive pole winding 0.24 Positive tab error 0.23 Abnormal cutting position of positive electrode 0.10 Negative electrode cutting position 0.10 Negative tab error 0.06

[0043] Light-GBM is a model based on the gradient boosting algorithm and the Boosting integrated learning strategy. Based on the residual output of each trained gradient boosting base model, a large number of tree models are iteratively trained in series to reduce the loss function until the set stopping condition is reached.

[0044] Compared with XGBoost, Light-GBM has the following advantages:

[0045] Histogram-based decision tree algorithm: Use binning and discrete processing for features to improve computing efficiency and reduce storage

[0046] Gradient-based One-Side Sampling (GOSS): GOSS can reduce the number of data instances with small gradients and only use data with high gradients to calculate information gain, which saves time and space compared to XGBoost traversing all eigenvalues.

[0047] Exclusive Feature Bundling (EFB): Use EFB to bundle mutually exclusive features into one feature to achieve dimensionality reduction

[0048] Leaf-wise growth strategy with depth limit: Most GBDT tools use an inefficient level-wise decision tree growth strategy, which still splits leaves with low splitting gain, increasing time consumption. Light-GBM uses a leaf-wise growth algorithm with depth limit.

[0049] Based on the above characteristic input factors, the capacity prediction model is obtained by constructing the relationship between the characteristics and the capacity compensation capacity.

[0050] For the parameters that affect the model's key parameters L1, L2, learning_rate, max_depth, n_estimators, etc., grid adjustment is used to obtain the adjustment parameters based on the mse indicator, as shown in the following Light-GBM optimization parameter table:

[0051] Light-GBM optimization parameters Optimization results Remark lambda_l1 0.014 L1 regularization coefficient lambda_l2 0.174 L2 regularization coefficient learning_rate 0.136 Learning Rate max_depth 12 Tree Depth n_estimators 20 Number of trees

[0052] DEA efficiency evaluation module (DEA) is a method for calculating the relative efficiency of decision-making units with multiple inputs and multiple outputs. This method does not require specific function forms and parameter weights. It calculates the weights of each feature based on linear programming to avoid human subjective judgment.

[0053] It has the following features:

[0054] The less data input, the better; the more data output, the better;

[0055] Characteristic indicators cannot be negative, and negative indicators need to be processed;

[0056] Feature inputs may not be independent;

[0057] No dimension normalization is required;

[0058] The number of decision-making units DMU>15. Too few DMUs will make it impossible to find the frontier.

[0059] DEA Model:

[0060] Common DEA models include CCR model, BCC model, super efficiency model, and SBM model.

[0061] Among them, CCR is a constant returns to scale model, that is, input / output changes linearly in equal proportion.

[0062] The BCC model is a variable returns to scale model, that is, the input / output does not change in proportion, and the same input can obtain more output.

[0063] The super-efficiency DEA model removes the DMU0 constraint of the CCR model itself, so that the DMU efficiency value is >1, which can effectively distinguish decision-making units.

[0064] SBM is mainly aimed at the traditional model (CCR BCC) that cannot handle the output factors containing non-desirable output factors (opposite to the desired output, such as pollution).

[0065] For a detailed derivation of DEA, please refer to the reference "Wei Quanling: Data Envelopment Analysis". The format after dualization is as follows:

[0066] Common BCC models are as follows:

[0067]

[0068] The CCR model is similar to the BCC model, except that there are fewer constraints.

[0069] in:

[0070] θ is the efficiency coefficient,

[0071] λ r is the slack adjustment amount and weight coefficient of the i-th input and j-th output.

[0072] X i , Y j is the i-th input and j-th output variable.

[0073] Efficiency calculation process

[0074] Based on module 1 characteristic factors and wide table data, they are used as inputs for the efficiency evaluation calculation module.

[0075] Input factors: various alarm quantities, startup time

[0076] Output factors: production capacity, yield

[0077] Efficiency calculation

[0078] CCR calculation is the comprehensive efficiency, BCC calculation is the pure technical efficiency, scale efficiency = comprehensive efficiency / pure technical efficiency

[0079] The specific calculation is shown in the DEA model section.

[0080] Technical efficiency: reflects the efficiency loss caused by input factors under the current input status.

[0081] Scale efficiency: reflects the efficiency loss caused by output factors under the current output status.

[0082] When technical efficiency = 1, it means that the decision-making unit is in the optimal scale state;

[0083] When technical efficiency is <1, it means that the decision-making unit has unreasonable investment;

[0084] When scale efficiency = 1, it means that the decision-making unit is in the optimal scale state;

[0085] When scale efficiency is <1, it means that the decision-making unit has unreasonable output;

[0086]

[0087] Machines with lower overall efficiency: 3#, 10#, 4#, 9#

[0088] Machines with low technical efficiency: 3#, 8#, 10# need to reduce the input alarm volume

[0089] Machines with low output efficiency: 3#, 10#, 4#, need to increase output

[0090] Slack variable adjustment

[0091] The CCR and BCC model efficiencies < 1 indicate that the technology is invalid;

[0092] >1 and slack variables Equal to 0 is strong and effective;

[0093] >1 and slack variables If it is not 0, it is weakly effective;

[0094] As shown in the above data wide table, the efficiency results only give the adjustment direction, and the relaxation variables again give the specific improvement adjustment amount to quantify the machine differences.

[0095] 4. Machine adjustment amount:

[0096] Input reduction adjustment factor: Negative electrode material defective tape 31, positive electrode cutting abnormality 1.62

[0097] Output increase adjustment amount: Total production 417, good products 402.

[0098] The present invention provides a winding machine capacity alarm efficiency evaluation method, which uses statistical means to analyze the relationship between the number of alarms in each process step, the downtime and the capacity and yield, locates the top factors affecting the capacity and yield, assists the business in outputting a maintenance list, and uses the DEA efficiency evaluation method to assist the business in outputting a maintenance list.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A winding machine production capacity alarm efficiency evaluation method, characterized in that: include: The alarm screening module determines the main influencing factors of single-volume NG, obtains the corresponding influencing factor data, cleans the obtained data, and uses the influencing factors as alarm factors and the production capacity as the output prediction quantity, thereby establishing a data wide table with input-output relationship; based on the importance of the lightgbm tree model, the characteristic factors in the data wide table are screened for features with high importance, and MSE is used as the basis for node splitting. In order to evaluate the quantitative relationship between the importance of the screened factors and the production capacity, the input-output relationship is fitted based on Light-GBM, and then the factor importance ranking is given under the premise of ensuring the fitting accuracy; DEA efficiency evaluation module, based on influencing factors and wide table data as the input of the efficiency evaluation calculation module, CCR calculation obtains the comprehensive efficiency, BCC calculation obtains the technical efficiency, and scale efficiency = comprehensive efficiency / technical efficiency, where technical efficiency: reflects the efficiency loss caused by input factors under the current input status; scale efficiency: reflects the efficiency loss caused by output factors under the current output status.

2. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The influencing factors include: abnormal positive electrode cutting position, positive electrode ear error, positive electrode single-pole winding, abnormal negative electrode cutting position, negative electrode material cutting, negative electrode ear error, negative electrode edge alarm, diaphragm hot hole error, positive electrode insertion length alarm, negative electrode insertion length alarm, positive electrode covering tape 2 error, positive electrode covering tape 2 contact check, poor CCD detection, abnormal rear winding length, and poor collection machine.

3. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The data cleaning process mainly deletes records of downtime greater than 3 hours during the maintenance period and abnormal production capacity records.

4. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The calculation formula of the MSE is: Among them, N is the number of samples of the current node, y i is the current node value, node prediction value MSE=(y i -mean) 2 / N.

5. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The screened factors are mainly selected, and the top 6 factors after MSE calculation are output as important factors.

6. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The Light-GBM is based on characteristic input factors and obtains a capacity prediction model by constructing the relationship between characteristics and capacity division compensation capacity. The parameter adjustment parameters are obtained based on the mse indicator by using grid adjustment to target key parameters affecting the model.

7. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The BCC model is as follows: The CCR model is similar to the BCC model, except that there are fewer constraints. Where: θ is the efficiency coefficient, λ r is the slack adjustment amount and weight coefficient of the i-th input and j-th output, X i , Y j is the i-th input and j-th output variable.

8. A winding machine production capacity alarm efficiency evaluation method as claimed in claim 1, characterized in that: The meanings of the technical efficiency and scale efficiency are as follows: When technical efficiency = 1, it means that the decision-making unit is in the optimal scale state; When technical efficiency is <1, it means that the decision-making unit has unreasonable investment; When scale efficiency = 1, it means that the decision-making unit is in the optimal scale state; When the scale efficiency is <1, it means that the decision-making unit has unreasonable output.

Images