CN114239378A - Injection molding product size prediction method based on custom LightGBM model loss - Google Patents
Injection molding product size prediction method based on custom LightGBM model loss Download PDFInfo
- Publication number
- CN114239378A CN114239378A CN202111337443.6A CN202111337443A CN114239378A CN 114239378 A CN114239378 A CN 114239378A CN 202111337443 A CN202111337443 A CN 202111337443A CN 114239378 A CN114239378 A CN 114239378A
- Authority
- CN
- China
- Prior art keywords
- injection molding
- historical
- size
- data
- prediction
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Medical Informatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Computer Hardware Design (AREA)
- Geometry (AREA)
- Injection Moulding Of Plastics Or The Like (AREA)
Abstract
The invention discloses an injection molding product size prediction method based on custom LightGBM model loss, which comprises the following steps: s1, acquiring historical SPC data of the injection molding process and historical high-frequency sampling data of the mold filling process state detection sensor and preprocessing the historical SPC data and the historical high-frequency sampling data; s2, extracting high-frequency characteristics from historical high-frequency sampling data of the mold filling process state detection sensor, and splicing the high-frequency characteristics with historical SPC data of the injection molding product forming process to form a data set; s3, performing variance filtering on the data set; s4, dividing the data set into a training set and a testing set; s5, training the custom-lost LightGBM model by utilizing the training set, verifying the prediction effect on the testing set, and finally using the LightGBM model for size prediction of the injection molding product. Compared with the traditional off-line detection method, the method greatly improves the detection efficiency and realizes the full detection of the product.
Description
Technical Field
The invention belongs to the technical field of machine learning, and particularly relates to an injection molding product size prediction method based on custom LightGBM model loss.
Background
The quality problem of the product is one of the most concerned problems in the production and manufacturing process, is a key ring in competition for market strategy of enterprises, and is the core power for determining the competitive advantages of the enterprises. The traditional product quality control mode is basically based on the off-line detection of the product quality, but the off-line detection belongs to post-inspection and has time lag, and because the limitation of the detection cost and the detection speed are difficult to match with the production efficiency of the product, the full inspection of the product is difficult to realize, and the reliability of the product quality cannot be guaranteed.
In recent years, with the continuous development of technologies such as big data and artificial intelligence, the quality prediction technology based on data driving is gradually mature, so that the real-time performance and the comprehensiveness of product inspection are met, and the management and control modes of enterprises on product quality are changed. At present, machine learning algorithms based on random forests, support vector machines, XGboost, LightGBM and the like and deep learning algorithms based on neural networks and the like are widely applied to quality prediction of products, but no solution is provided for the problems of missing judgment and erroneous judgment existing in the quality prediction, so that difficulties and challenges are brought to the final product quality judgment problems, such as the problems of size missing judgment and erroneous judgment in injection molding product prediction.
Disclosure of Invention
The invention mainly aims to overcome the defects and shortcomings of the prior art, provides an injection molding product size prediction method based on user-defined LightGBM model loss, and solves the problems of missed judgment and erroneous judgment in injection molding product size prediction in the prior art.
In order to achieve the purpose, the invention adopts the following technical scheme:
an injection molding product size prediction method based on custom LightGBM model loss comprises the following steps:
s1, acquiring historical SPC data of the injection molding process and historical high-frequency sampling data of the mold filling process state detection sensor and preprocessing the historical SPC data and the historical high-frequency sampling data;
s2, extracting high-frequency characteristics from historical high-frequency sampling data of the mold filling process state detection sensor, and splicing the high-frequency characteristics with historical SPC data of the injection molding product forming process to form a data set;
s3, performing variance filtering on the data set;
s4, dividing the data set into a training set and a testing set;
s5, training the custom-lost LightGBM model by utilizing the training set, verifying the prediction effect on the testing set, and finally using the LightGBM model for size prediction of the injection molding product.
Further, step S1 specifically includes:
s11, filling the mean value of the missing values in the historical SPC data of the injection molding process;
and S12, carrying out mean value filling on missing values existing in the historical high-frequency sampling data of the mold filling process state detection sensor.
Further, in step S2, the high frequency characteristic is a minimum value of each column of characteristics in the historical high frequency sampling data of the mold filling process state detection sensor.
Further, step S3 specifically includes:
s31, calculating the variance of each column of characteristics of the data set;
and S32, eliminating the features with the variance of zero.
Further, step S5 specifically includes:
s51, introducing a penalty factor into a loss function of the LightGBM;
s52, training the LightGBM model by using the training set, predicting the size of the injection molding product on the test set, counting the Mean Square Error (MSE), the rate of missing judgment and the rate of false judgment of the prediction result, and verifying the prediction result.
Further, the loss function is specifically:
where N is the number of samples, yiIs the true value of the ith sample,for the predicted value of the ith sample, Ω is the structural loss associated with the regularization term,the target penalty factor is obtained by the following values:
wherein, ω is1For a target penalty factor, ω, in the case of a false positive2For a target penalty factor in the case of a missed judgment, LL is a lower size limit, UL is an upper size limit, yiIs the true value of the ith sample,is the predicted value of the ith sample.
Further, the missing rate specifically is:
the rate of missed judgment is the number of missed judgment of the prediction samples/the total number of the prediction samples
The misjudgment rate is specifically as follows:
the misjudgment rate is the misjudgment number of the prediction samples/the total number of the prediction samples
The number of the missed samples is specifically predicted by the number of samples, wherein the real size value of the injection molding product is greater than the upper size limit UL or less than the lower size limit LL, but the predicted value is greater than or equal to the lower size limit LL and less than or equal to the upper size limit UL;
the number of misjudgments of the pre-judged samples is specifically the number of samples of which the real size value of the injection molding product is greater than or equal to the size lower limit LL and less than or equal to the size upper limit UL, but the predicted value is greater than the size upper limit UL or less than the size lower limit LL.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. compared with the traditional offline detection method, the online detection method based on machine learning greatly improves the detection efficiency, and realizes the full detection of the product.
2. Compared with the traditional offline detection method, the online detection method based on machine learning reduces the labor intensity of workers and effectively reduces the detection cost.
3. Compared with the existing machine learning online detection algorithm, the method introduces corresponding penalty factors into the loss function of the LightBGM model, reduces the conditions of missing judgment and misjudgment in the prediction result, and improves the comprehensive prediction performance of the model.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
fig. 2 is a comparison graph of the predicted size distribution of the missing-judged and erroneous-judged samples of the custom-lost LightGBM model and the LightGBM model in the embodiment.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Examples
As shown in fig. 1, the method for predicting the dimension of an injection molded article based on the loss of the custom LightGBM model of the present invention comprises the following steps:
s1, acquiring historical SPC data of the injection molding process and historical high-frequency sampling data of the mold filling process state detection sensor and preprocessing, wherein the preprocessing comprises the following substeps:
s11, filling the mean value of the missing values in the historical SPC data of the injection molding process;
and S12, carrying out mean value filling on missing values existing in the historical high-frequency sampling data of the mold filling process state detection sensor.
In this embodiment, the historical SPC data of the injection molding process and the historical high-frequency sampling data of the mold filling process state detection sensor in the fourth industrial big data innovation competition data set are used. The dimension of historical SPC data in the injection molding product forming process is (20553, 86), 20553 is the total number of the die times of the product, and 86 is the characteristic number. The dimensions of historical high-frequency sampling data of the mold filling process state detection sensor are (20533, 19, 1500 +/-200), 20553 is the total number of the mold times of the product, 19 is the number of characteristics, and 1500 +/-200 is the number of data records collected by the high-frequency document transmission device in each mold time.
S2, extracting high-frequency characteristics from historical high-frequency sampling data of the mold filling process state detection sensor, and splicing the high-frequency characteristics with historical SPC data of the injection molding product forming process to form a data set; the high-frequency characteristic is the minimum value of each column of characteristics in the historical high-frequency sampling data of the mold filling process state detection sensor.
In this embodiment, the dimension of the data set formed by splicing is (20553, 105).
S3, performing variance filtering on the data set, including:
s31, calculating the variance of each column of characteristics of the data set;
and S32, eliminating the features with the variance of zero.
S4, dividing the data set into a training set and a testing set; in this embodiment, the training set dimensions of the partition are (16600, 105), (3953, 105).
S5, training a custom-lost LightGBM model by utilizing a training set, verifying a prediction effect on a test set, and finally using the LightGBM model for size prediction of an injection molding product; the method specifically comprises the following steps:
s51, introducing a penalty factor into a loss function of the LightGBM; the loss function is specifically:
where N is the number of samples, yiIs the true value of the ith sample,for the predicted value of the ith sample, Ω is the structural loss associated with the regularization term,the target penalty factor is obtained by the following values:
wherein, ω is1For a target penalty factor, ω, in the case of a false positive2For a target penalty factor in the case of a missed judgment, LL is a lower size limit, UL is an upper size limit, yiIs the true value of the ith sample,is the predicted value of the ith sample.
In the present embodiment, ω1Value of 10, omega2The value is 10, the lower size limit LL is 199.925mm, and the upper size limit UL is 200.075.
S52, training a LightGBM model by using a training set, predicting the size of an injection molding product on a test set, counting the Mean Square Error (MSE), the rate of missing judgment and the rate of erroneous judgment of a prediction result, and verifying the prediction result; the rate of missed judgment is specifically as follows:
the rate of missed judgment is the number of missed judgment of the prediction samples/the total number of the prediction samples
The misjudgment rate is specifically as follows:
the misjudgment rate is the misjudgment number of the prediction samples/the total number of the prediction samples
The number of the missed samples is specifically predicted by the number of samples, wherein the real size value of the injection molding product is greater than the upper size limit UL or less than the lower size limit LL, but the predicted value is greater than or equal to the lower size limit LL and less than or equal to the upper size limit UL;
the number of misjudgments of the pre-judged samples is specifically the number of samples of which the real size value of the injection molding product is greater than or equal to the size lower limit LL and less than or equal to the size upper limit UL, but the predicted value is greater than the size upper limit UL or less than the size lower limit LL.
In this example, the model is trained using a training set, with the hyper-parameters of the model as follows: the boosting type boosting _ type is set to gbdt, the number of leaves of the tree num _ leaves is set to 25, the learning rate learning _ rate is set to 0.3, the number of CART n _ estimators is set to 42, the feature selection ratio feature _ fraction of the tree building is set to 0.9, the sample sampling ratio bagging _ fraction of the tree building is set to 0.8, and bagging _ fraction is set to 5 every k iterations.
The statistical results of mean square error MSE, missed rate and false rate are shown in table 1 below.
TABLE 1
From table 1, it can be seen that the dimension prediction method for the injection molded product based on the loss of the custom LightGBM model greatly improves the problems of missing judgment and misjudgment in the dimension prediction under the condition that the MSE is not very different, and compared with the traditional LightGBM model taking the mean square error as the loss function, the missing judgment rate is reduced by 46.8%, and the misjudgment rate is reduced by 24.4%.
In order to more intuitively show the comprehensive prediction performance improvement condition of the model after the user-defined loss function is used, the distribution conditions of all samples which are not judged and are judged wrongly in the prediction results of the two prediction models are visualized, and the results are shown in fig. 2.
As can be seen from FIG. 2, after the user-defined loss function is used for the sample data with the occurrence of the missing judgment and the erroneous judgment, the whole predicted size of the model is closer to the actual measured size, and the obtained model has better product size prediction performance. The LightGBM model is finally used for injection molded article dimension prediction.
It should also be noted that in this specification, terms such as "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (7)
1. An injection molding product size prediction method based on custom LightGBM model loss is characterized by comprising the following steps:
s1, acquiring historical SPC data of the injection molding process and historical high-frequency sampling data of the mold filling process state detection sensor and preprocessing the historical SPC data and the historical high-frequency sampling data;
s2, extracting high-frequency characteristics from historical high-frequency sampling data of the mold filling process state detection sensor, and splicing the high-frequency characteristics with historical SPC data of the injection molding product forming process to form a data set;
s3, performing variance filtering on the data set;
s4, dividing the data set into a training set and a testing set;
s5, training the custom-lost LightGBM model by utilizing the training set, verifying the prediction effect on the testing set, and finally using the LightGBM model for size prediction of the injection molding product.
2. The method of claim 1, wherein the step S1 specifically comprises:
s11, filling the mean value of the missing values in the historical SPC data of the injection molding process;
and S12, carrying out mean value filling on missing values existing in the historical high-frequency sampling data of the mold filling process state detection sensor.
3. The method of claim 1, wherein in step S2, the high frequency characteristic is a minimum value of each column of characteristics in the historical high frequency sampling data of the sensor for detecting the status of the mold filling process.
4. The method of claim 1, wherein the step S3 specifically comprises:
s31, calculating the variance of each column of characteristics of the data set;
and S32, eliminating the features with the variance of zero.
5. The method of claim 1, wherein the step S5 specifically comprises:
s51, introducing a penalty factor into a loss function of the LightGBM;
s52, training the LightGBM model by using the training set, predicting the size of the injection molding product on the test set, counting the Mean Square Error (MSE), the rate of missing judgment and the rate of false judgment of the prediction result, and verifying the prediction result.
6. The method of claim 5, wherein the loss function is specifically as follows:
where N is the number of samples, yiIs the true value of the ith sample,for the predicted value of the ith sample, Ω is the structural loss associated with the regularization term,the target penalty factor is obtained by the following values:
7. The method of claim 5, wherein the missing rate is specifically:
the rate of missed judgment is the number of missed judgment of the prediction samples/the total number of the prediction samples
The misjudgment rate is specifically as follows:
the misjudgment rate is the misjudgment number of the prediction samples/the total number of the prediction samples
The number of the missed samples is specifically predicted by the number of samples, wherein the real size value of the injection molding product is greater than the upper size limit UL or less than the lower size limit LL, but the predicted value is greater than or equal to the lower size limit LL and less than or equal to the upper size limit UL;
the number of misjudgments of the pre-judged samples is specifically the number of samples of which the real size value of the injection molding product is greater than or equal to the size lower limit LL and less than or equal to the size upper limit UL, but the predicted value is greater than the size upper limit UL or less than the size lower limit LL.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111337443.6A CN114239378A (en) | 2021-11-09 | 2021-11-09 | Injection molding product size prediction method based on custom LightGBM model loss |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111337443.6A CN114239378A (en) | 2021-11-09 | 2021-11-09 | Injection molding product size prediction method based on custom LightGBM model loss |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114239378A true CN114239378A (en) | 2022-03-25 |
Family
ID=80749177
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111337443.6A Pending CN114239378A (en) | 2021-11-09 | 2021-11-09 | Injection molding product size prediction method based on custom LightGBM model loss |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114239378A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115760782A (en) * | 2022-11-16 | 2023-03-07 | 华南理工大学 | In-mold labeling offset defect identification method based on machine vision |
-
2021
- 2021-11-09 CN CN202111337443.6A patent/CN114239378A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115760782A (en) * | 2022-11-16 | 2023-03-07 | 华南理工大学 | In-mold labeling offset defect identification method based on machine vision |
CN115760782B (en) * | 2022-11-16 | 2023-06-16 | 华南理工大学 | Machine vision-based in-mold labeling offset defect identification method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104914723B (en) | Industrial process soft-measuring modeling method based on coorinated training partial least square model | |
CN107679715A (en) | A kind of electric energy meter comprehensive error process merit rating method and evaluation system based on SPC | |
CN103488561B (en) | A kind of kNN fault detection method of online upgrading master sample model | |
CN111768000A (en) | Industrial process data modeling method for online adaptive fine-tuning deep learning | |
CN107846472A (en) | The fleet anomaly detection method of extensive power transmission and transforming equipment Monitoring data flow | |
CN109298225B (en) | Automatic identification model system and method for abnormal state of voltage measurement data | |
CN114239378A (en) | Injection molding product size prediction method based on custom LightGBM model loss | |
CN105844334B (en) | A kind of temperature interpolation method based on radial base neural net | |
CN109783898A (en) | A kind of intelligent optimization method of injection molding manufacturing technique parameter | |
CN113919601A (en) | Resin process prediction method and device based on product performance and process data model | |
CN108334898A (en) | A kind of multi-modal industrial process modal identification and Fault Classification | |
CN112305388B (en) | On-line monitoring and diagnosing method for insulation partial discharge faults of generator stator winding | |
CN105699437A (en) | Rapid nondestructive distinguishing method of Chinese walnuts with different fresh degrees based on electronic nose | |
CN109144028A (en) | A kind of rectifying column efficiency deterioration detecting | |
CN111461192B (en) | River channel water level flow relation determination method based on multi-hydrological station linkage learning | |
CN113723844A (en) | Low-voltage transformer area theoretical line loss calculation method based on ensemble learning | |
CN117370744A (en) | Dynamic cleaning method and system for abnormal power consumption data of power consumer | |
Wang et al. | A novel sensing feature extraction based on mold temperature and melt pressure for plastic injection molding quality assessment | |
CN112164144B (en) | Casting three-dimensional model classification method combining D2 operator and normal operator | |
CN107507885A (en) | Manufacture of solar cells process monitoring method based on multichannel sensor data | |
CN114372640A (en) | Wind power prediction method based on fluctuation sequence classification correction | |
Gao et al. | Traffic flow forecasting based on wavelet neural network optimized by GA | |
CN114186844A (en) | Method and device for identifying electricity stealing clients | |
CN109902378B (en) | Complex industrial process low-cost modeling method based on multi-model migration and BMA theory | |
CN109947082A (en) | A kind of process monitoring method based on collection nucleation independent component analysis model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |