CN115129019A - Training method of production line fault analysis model and production line fault analysis method - Google Patents

Training method of production line fault analysis model and production line fault analysis method Download PDF

Info

Publication number
CN115129019A
CN115129019A CN202211060783.3A CN202211060783A CN115129019A CN 115129019 A CN115129019 A CN 115129019A CN 202211060783 A CN202211060783 A CN 202211060783A CN 115129019 A CN115129019 A CN 115129019A
Authority
CN
China
Prior art keywords
data
fault analysis
analysis model
production line
training
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
Application number
CN202211060783.3A
Other languages
Chinese (zh)
Inventor
令狐彬
胡炳彰
许�鹏
周璠
卫峥
高磊
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Hefei Zhongke Dihong Automation Co ltd
Original Assignee
Hefei Zhongke Dihong Automation Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Hefei Zhongke Dihong Automation Co ltd filed Critical Hefei Zhongke Dihong Automation Co ltd
Priority to CN202211060783.3A priority Critical patent/CN115129019A/en
Publication of CN115129019A publication Critical patent/CN115129019A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Programme-control systems
    • G05B19/02—Programme-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]
    • G05B19/41875—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM] characterised by quality surveillance of production
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B2219/00—Program-control systems
    • G05B2219/30—Nc systems
    • G05B2219/31—From computer integrated manufacturing till monitoring
    • G05B2219/31357—Observer based fault detection, use model

Landscapes

  • Engineering & Computer Science (AREA)
  • General Engineering & Computer Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a training method of a production line fault analysis model and a production line fault analysis method, wherein the training method comprises the following steps: acquiring a data set, wherein the data set comprises a training set and a test set; constructing a fault analysis model; and training the fault analysis model by using the data in the training set to minimize the loss function, and testing the trained fault analysis model by using the data in the testing set to obtain a final fault analysis model. Therefore, the training method can improve the recognition capability of the fault analysis model on the defect mode corresponding to the fault type, so that the fault type generated on the processing production line can be better predicted.

Description

Training method of production line fault analysis model and production line fault analysis method
Technical Field
The invention relates to the technical field of computer machine learning and artificial intelligence, in particular to a training method of a production line fault analysis model, a production line fault analysis method and electronic equipment.
Background
The AOI (Automated Optical Inspection) is an apparatus for inspecting common defects encountered in the production line based on the Optical principle, and is an Automated visual Inspection in the processes of PCB (Printed Circuit Board), LCD (Liquid Crystal Display), transistor manufacturing or product processing, which is suitable for mass production and can be implemented in many stages in the manufacturing process. When the AOI device detects that a specific device on the production line is under a specific environment and the number of specific defects is increased, it means that the specific device on the production line is out of order. In traditional production, an engineer or AOI management equipment monitors the quantity and the condition of defect generation, and the engineer inspects each link and equipment of a production line according to experience to judge the link and equipment of fault generation. However, the traditional fault analysis method depends on the experience of engineers, has poor timeliness and is not beneficial to production.
In addition, when the common linear model is used for processing the feature data, all the features are considered independently, the correlation between the features is not considered, and the relation between the defects and the faults cannot be well established. In practice, however, there may be some correlation between features.
However, at present, relatively few researches on fault analysis by using AOI data are conducted, and feature extraction of AOI data is not sufficient, so that the fault identification capability is still low, and a place for improvement exists.
Disclosure of Invention
The present invention is directed to solving, at least in part, one of the technical problems in the related art. Therefore, a first objective of the present invention is to provide a method for training a production line fault analysis model, which can improve the recognition capability of the fault analysis model for the fault mode corresponding to the fault type, so as to better predict the fault type generated on the processing production line.
The second purpose of the invention is to provide a production line fault analysis method.
A third object of the invention is to propose an electronic device.
In order to achieve the above object, an embodiment of a first aspect of the present invention provides a training method for a production line fault analysis model, where the training method includes the following steps: acquiring a data set, wherein the data set comprises a training set and a testing set, and the data in the data set comprises defect data detected by automatic optical detection equipment on a production line, environment data of the environment where the production line is located and corresponding fault category labels; constructing a fault analysis model, wherein the fault analysis model comprises a factorization machine module, a deep neural network module, a feature selection module and a classification module, input data of the factorization machine module and input data of the deep neural network module are both the defect data and the environment data, and correspondingly output low-order features and high-order features, the feature selection module is used for performing feature selection on the low-order features and the high-order features, and the classification module is used for obtaining probability distribution of fault categories according to the selected features; and training the fault analysis model by using the data in the training set to minimize a loss function, and testing the trained fault analysis model by using the data in the testing set to obtain a final fault analysis model.
According to the training method of the production line fault analysis model provided by the embodiment of the invention, the recognition capability of the fault analysis model on the defect mode corresponding to the fault type can be improved, so that the fault type generated on the processing production line can be better predicted.
In addition, the training method of the production line fault analysis model provided by the embodiment of the invention also has the following additional technical characteristics:
further, the acquiring the data set includes: counting defect data, environment data and fault category labels in a plurality of time periods, wherein a defect data set is obtained for each time period
Figure 235170DEST_PATH_IMAGE001
Environment data collection
Figure 166830DEST_PATH_IMAGE002
And a corresponding label for a category of failure,
Figure 536631DEST_PATH_IMAGE003
as the number of the j-th type defects,
Figure 692806DEST_PATH_IMAGE004
as to the number of defect classes,
Figure 489861DEST_PATH_IMAGE005
for the i-th type of environment data,
Figure 161014DEST_PATH_IMAGE006
the number of categories of environmental data.
According to some embodiments of the invention, prior to inputting data in the data set to the fault analysis model, statistical data for each time period is preprocessed as follows: normalizing various types of defect data in the defect data set to obtain a normalized defect data set
Figure 68927DEST_PATH_IMAGE007
Wherein
Figure 79608DEST_PATH_IMAGE008
normalized number of j-th class defects; carrying out thermal independent coding on various environmental data in the environmental data set to obtain a thermal independent coding set of the environmental data
Figure 47564DEST_PATH_IMAGE009
Wherein
Figure 956745DEST_PATH_IMAGE010
encoding the thermal independent code of the ith type environment data; carrying out Dense embedding processing on the thermal independent coding set to obtain a low-dimensional Dense representation set
Figure 668350DEST_PATH_IMAGE011
(ii) a Will be provided with
Figure 799117DEST_PATH_IMAGE012
And
Figure 291DEST_PATH_IMAGE013
splicing to obtain the characteristic vector of the input data
Figure 318140DEST_PATH_IMAGE014
Is recorded as
Figure 833435DEST_PATH_IMAGE015
Wherein
Figure 818708DEST_PATH_IMAGE016
is the number of data in X.
Further, in each training period, selecting an input data feature vector of a time period for training, wherein the factorization module obtains a low-order feature y of the input data feature vector X according to the following formula FM ：
Figure 941516DEST_PATH_IMAGE017
Wherein
Figure 746661DEST_PATH_IMAGE018
for the p-th input data feature in X,
Figure 800068DEST_PATH_IMAGE019
and
Figure 639848DEST_PATH_IMAGE020
are parameters of the factorizer module and,
Figure 182824DEST_PATH_IMAGE021
is the product of two k-dimensional vectors.
According to some embodiments of the invention, the deep neural network module comprises a plurality of series-connected feedforward neural network modules, each of the feedforward neural networksThe deep neural network module sequentially comprises a full connection layer, a ReLU activation function and a Dropout layer, and obtains high-order features y of input data feature vectors X through the following formula FNN ：
Figure 740845DEST_PATH_IMAGE022
。
According to some embodiments of the invention, the feature selection module is a feature selection module fusing attention mechanisms, the feature selection module deriving the attention selection feature by:
Figure 332363DEST_PATH_IMAGE023
wherein
Figure 292229DEST_PATH_IMAGE024
the feature is selected for the purpose of said attention,
Figure 756839DEST_PATH_IMAGE025
，
Figure 536577DEST_PATH_IMAGE026
，
Figure 931786DEST_PATH_IMAGE027
，
Figure 746158DEST_PATH_IMAGE028
in order to splice the feature vectors, the feature vectors are,
Figure 630937DEST_PATH_IMAGE029
、
Figure 897971DEST_PATH_IMAGE030
and
Figure 96871DEST_PATH_IMAGE031
is a preset weight matrix and is used for carrying out weight adjustment,
Figure 765750DEST_PATH_IMAGE032
the dimension of the vector matrix K.
According to some embodiments of the invention, the classification module comprises a fully connected layer and a softmax layer, the classification module deriving the probability distribution of the fault classes according to the attention-selection features by:
Figure 575092DEST_PATH_IMAGE033
wherein Z is the probability distribution of the fault category.
According to some embodiments of the invention, the loss function of the fault analysis model is calculated by
Figure 329422DEST_PATH_IMAGE034
：
Figure 332013DEST_PATH_IMAGE035
Wherein, in the process,
Figure 855398DEST_PATH_IMAGE036
is as follows
Figure 81980DEST_PATH_IMAGE037
A fault-like label, T is the number of fault category labels,
Figure 323606DEST_PATH_IMAGE038
is as follows
Figure 864308DEST_PATH_IMAGE037
Probability of class failure.
In order to achieve the above object, a second aspect of the present invention provides a method for analyzing a production line fault, where the method includes the following steps: acquiring data to be analyzed, wherein the data to be analyzed comprises defect data and environmental data on a production line; and carrying out fault analysis on the production line according to the data to be analyzed by utilizing the fault analysis model obtained by the training method of the production line fault analysis model.
According to the production line fault analysis method provided by the embodiment of the invention, the fault analysis model obtained by the training method of the production line fault analysis model can be used for improving the accuracy of fault analysis.
In order to achieve the above object, a third aspect of the present invention provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory, where the computer program, when executed by the processor, implements the method for training the line fault analysis model or implements the method for line fault analysis.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
FIG. 1 is a schematic flow diagram of a method of training a production line fault analysis model in accordance with some embodiments of the invention;
FIG. 2 is a schematic flow diagram of a method of line fault analysis in accordance with some embodiments of the invention;
fig. 3 is a schematic structural diagram of an electronic device according to some embodiments of the invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the accompanying drawings are illustrative and intended to explain the present invention and should not be construed as limiting the present invention.
The following describes a training method of a production line fault analysis model and a production line fault analysis method according to an embodiment of the present invention with reference to fig. 1 to 3.
FIG. 1 is a schematic flow diagram of a method of training a production line fault analysis model according to some embodiments of the invention. As shown in FIG. 1, the training method of the production line fault analysis model comprises the following steps:
s101, acquiring a data set, wherein the data set comprises a training set and a testing set, and data in the data set comprises defect data detected by automatic optical detection equipment on a production line, environment data of the environment where the production line is located and corresponding fault category labels.
Specifically, data set A includes training set A1 and test set A2. Acquiring a data set may include: counting defect data, environment data and fault category labels in a plurality of time periods, wherein a defect data set is obtained for each time period
Figure 242200DEST_PATH_IMAGE039
Environment data collection
Figure 655995DEST_PATH_IMAGE002
And a corresponding fault category label, and,
Figure 119337DEST_PATH_IMAGE003
as the number of the j-th type defects,
Figure 463731DEST_PATH_IMAGE004
as to the number of defect classes,
Figure 758446DEST_PATH_IMAGE040
for the i-th type of environment data,
Figure 530093DEST_PATH_IMAGE041
the number of categories of environmental data.
Specifically, defect data (such as appearance, scratch, measurement data, size, angle and the like) detected by automatic optical detection equipment on the production line in a plurality of time periods, environmental data (such as operating parameters of the equipment, environmental temperature, humidity and the like which are related to the operating state of the equipment) of the environment where the production line is located and fault category labels are counted. Taking t minutes as a time period, and obtaining a defect data set aiming at each time period
Figure 480732DEST_PATH_IMAGE039
Environmental data set
Figure 363237DEST_PATH_IMAGE002
And a corresponding fault category label, wherein,
Figure 528770DEST_PATH_IMAGE003
as the number of the j-th type defects,
Figure 471318DEST_PATH_IMAGE004
the number of defect classes detected for automated optical inspection equipment,
Figure 909253DEST_PATH_IMAGE040
is the (i) th type of environment data,
Figure 595449DEST_PATH_IMAGE006
the number of categories of environmental data. It should be noted that, an engineer checks the fault condition of the processing equipment in each time period, records that the processing equipment is normal or a corresponding fault category in the time period, and forms a fault category label.
S102, a fault analysis model is constructed, wherein the fault analysis model comprises a factorization machine module, a deep neural network module, a feature selection module and a classification module, input data of the factorization machine module and input data of the deep neural network module are both defect data and environment data, low-order features and high-order features are correspondingly output, the feature selection module is used for carrying out feature selection on the low-order features and the high-order features, and the classification module is used for obtaining probability distribution of fault categories according to the selected features.
Specifically, the data in the data set comprises defect data detected by automatic optical detection equipment on a production line, environment data of the environment where the production line is located, an input factor decomposition (Factorization Machines) model and a deep neural network module, low-order features and high-order features are correspondingly output, feature selection is performed on the low-order features and the high-order features through a feature selection module, and finally probability distribution of fault categories is obtained through a classification module according to the selected features.
As one example, before inputting the data in the data set to the fault analysis model, the statistical data for each time period is preprocessed as follows: in a butt jointNormalizing various defect data in the defect data set to obtain a normalized defect data set
Figure 864757DEST_PATH_IMAGE007
Wherein
Figure 978206DEST_PATH_IMAGE008
normalized number of j-th class defects; carrying out thermal independent coding on various environmental data in the environmental data set to obtain a thermal independent coding set of the environmental data
Figure 903437DEST_PATH_IMAGE009
Wherein
Figure 127745DEST_PATH_IMAGE010
encoding the thermal independent code of the ith type environment data; carrying out Dense embedding processing on the thermal independent coding set to obtain a low-dimensional Dense representation set
Figure 267870DEST_PATH_IMAGE042
(ii) a Will be provided with
Figure 286642DEST_PATH_IMAGE012
And
Figure 699169DEST_PATH_IMAGE013
splicing to obtain the characteristic vector of the input data
Figure 727167DEST_PATH_IMAGE043
Is marked as
Figure 971067DEST_PATH_IMAGE015
Wherein
Figure 426319DEST_PATH_IMAGE016
is the number of data in X.
In particular, for defect data sets
Figure 60563DEST_PATH_IMAGE039
Number of various types of defects inPerforming normalization processing to obtain a normalized defect data set
Figure 436793DEST_PATH_IMAGE007
Wherein
Figure 738461DEST_PATH_IMAGE008
and the number of j-th defects is normalized, so that the convergence speed and stability of training are improved and the influence of extreme values and abnormal values is reduced by performing normalization processing on various types of defect data in the defect data set. The environmental data of the environment in which the production line is located can be divided into discrete class data and numerical class data. If the environmental data of the environment where the production line is located is discrete data, the environmental data is aggregated
Figure 364615DEST_PATH_IMAGE002
The various environmental data in the system are subjected to one-hot encoding (namely one-hot encoding) to obtain a hot-single encoding set of the environmental data
Figure 486154DEST_PATH_IMAGE009
Wherein
Figure 918273DEST_PATH_IMAGE010
and carrying out thermal independent coding on the ith type of environment data, and converting the category characteristics into vector representation which can be processed by a deep learning model by carrying out thermal independent coding on various types of environment data in the environment data set. Encoding a set of hot unique codes
Figure 74448DEST_PATH_IMAGE044
Performing Dense embedding (namely embedding) processing on (namely the high-dimensional sparse environment representation set) to obtain a low-dimensional Dense representation set
Figure 871502DEST_PATH_IMAGE042
. Finally, the normalized defect data is collected
Figure 480338DEST_PATH_IMAGE007
As a numerical feature vector, the lower dimensionDense representation collections
Figure 201301DEST_PATH_IMAGE011
As dense numerical value feature vectors, splicing the two to obtain input data feature vectors
Figure 211982DEST_PATH_IMAGE045
Is marked as
Figure 179938DEST_PATH_IMAGE015
Wherein
Figure 276070DEST_PATH_IMAGE016
is the number of data in X. Therefore, the defect data are normalized, the environment data are subjected to the independent hot coding, the result of the independent hot coding is subjected to the embedding processing, the normalized result and the embedding processing result are spliced, the data are processed into the numerical value characteristic vector, and the subsequent model processing effect is improved.
If the environmental data of the environment in which the production line is located is numerical data such as environmental temperature, humidity, and the like, normalization processing is performed on the various environmental data in the environmental data set to obtain a normalized environmental data set, and then the normalized defect data set and the normalized environmental data set are used as numerical feature vectors and are spliced to obtain input data feature vectors.
Further, in each training cycle, an input data feature vector X of a time period is selected for training.
In this example, the factorizer module is formulated by
Figure 49991DEST_PATH_IMAGE017
Obtaining the low-order feature y of the input data feature vector X FM Wherein
Figure 915179DEST_PATH_IMAGE018
for the p-th input data feature in X,
Figure 54036DEST_PATH_IMAGE016
is the amount of data in X and,
Figure 637464DEST_PATH_IMAGE019
and
Figure 965808DEST_PATH_IMAGE020
are the parameters of the factoring machine module,
Figure 685503DEST_PATH_IMAGE021
is the product of two k-dimensional vectors. It should be noted that the building factor decomposition module learns the cross features in the form of dot products and hidden vectors, and generally learns the cross features of the second order. Therefore, the cross relation among the input features is extracted by the factor decomposition machine module, the feature combination problem under sparse data is solved, the prediction complexity is linear, the method has good universality on continuous and discrete features, and the feature cross which never or rarely occurs in a training set can be better learned.
In this example, the deep Neural network module comprises a plurality of series-connected Feed-forward Neural network modules (Feed-forward Neural Networks), each of which comprises a fully-connected layer, a ReLU activation function, and a Dropout layer (hidden layer) in this order, and the deep Neural network module is formulated by the formula
Figure 995261DEST_PATH_IMAGE046
Obtaining high-order characteristic y of input data characteristic vector X FNN . Therefore, by constructing the deep neural network module, the cross relation of high-order features can be extracted, and the accuracy of fault classification of the fault analysis model is improved.
It should be noted that, the factorization machine module and the deep neural network module share the input data feature vector X, so that the low-order and high-order feature intersections can be simultaneously learned from the original features, and end-to-end training can be performed.
In this example, the feature selection module is a feature selection module that fuses attention mechanisms, the feature selection module being formulated by a formula
Figure 65986DEST_PATH_IMAGE023
Attention selection features are derived in which, among other things,
Figure 181709DEST_PATH_IMAGE047
the features are selected for the purpose of attention,
Figure 21489DEST_PATH_IMAGE025
，
Figure 502149DEST_PATH_IMAGE026
，
Figure 60169DEST_PATH_IMAGE048
，
Figure 464737DEST_PATH_IMAGE028
in order to splice the feature vectors,
Figure 159023DEST_PATH_IMAGE029
、
Figure 76164DEST_PATH_IMAGE049
and
Figure 855901DEST_PATH_IMAGE031
is a preset weight matrix and is used for carrying out weight adjustment,
Figure 313427DEST_PATH_IMAGE032
the dimension of the vector matrix K.
Specifically, the low-order features of the input data feature vector X output by the factorizer module are combined
Figure 127799DEST_PATH_IMAGE017
And the high-order features of the input data feature vector X output by the deep neural network module
Figure 950262DEST_PATH_IMAGE022
Splicing to obtain splicing characteristic vector
Figure 217295DEST_PATH_IMAGE028
. Using a formula
Figure 243893DEST_PATH_IMAGE025
，
Figure 912772DEST_PATH_IMAGE026
，
Figure 906136DEST_PATH_IMAGE048
Constructing Query, Key and Value matrixes which are respectively recorded as Q, K and V, calculating a vector matrix Q and a vector matrix K to obtain a similarity weight, and normalizing the weight to obtain a formula
Figure 660465DEST_PATH_IMAGE050
Wherein
Figure 459794DEST_PATH_IMAGE051
the calculated attention weight matrix for vector matrix Q and vector matrix K,
Figure 983179DEST_PATH_IMAGE032
the dimension of the vector matrix K. Finally, attention weight matrix
Figure 413023DEST_PATH_IMAGE051
Carrying out weighted summation with the vector matrix V to obtain attention selection characteristics
Figure 654649DEST_PATH_IMAGE023
. Therefore, the feature selection module fusing the attention mechanism is constructed by taking the spliced feature vector obtained from the factorization machine module and the deep neural network module as an original input and projecting the spliced feature vector to different high-dimensional vector spaces through Q, K and V matrixes and calculating the attention weight matrix, so that the features with high attention weight enhancement and fault correlation can be calculated, and the irrelevant features can be weakened.
In this example, the classification module comprises a fully connected layer and a softmax layer, and the classification module is formulated by formula
Figure 742822DEST_PATH_IMAGE033
And obtaining the probability distribution of the fault category according to the attention selection characteristics, wherein Z is the probability distribution of the fault category. Therefore, low-order features and high-order features of the defect data and the environment data are respectively extracted through the factorization machine module and the deep neural network module to obtain cross feature combination and implicit feature association, proper features are selected from the low-order features and the high-order features for prediction through the feature selection module integrating the attention mechanism, and probability distribution of fault categories is obtained through the classification module according to the selected features, so that more accurate fault analysis is achieved.
S103, training the fault analysis model by using the data in the training set to enable a loss function to be minimum, and testing the trained fault analysis model by using the data in the testing set to obtain a final fault analysis model.
Specifically, the training set A1 comprises defect data and environment data of multiple time periods, the defect data and the environment data in each time period are preprocessed to obtain an input data feature vector X, the input data feature vector X is input into a fault analysis model to predict fault types, and the fault analysis model is trained by using an Adam optimizer, so that a loss function is obtained
Figure 386293DEST_PATH_IMAGE034
To a minimum. Thus, by utilizing the Adam optimizer, the ability to select more beneficial features among the input features is adaptively learned. And testing the trained fault analysis model by using the data in the test set A2 so as to obtain a final fault analysis model, wherein the final fault analysis model can be used for fault analysis and positioning of a processing production line. Wherein, can pass through the formula
Figure 721459DEST_PATH_IMAGE035
Calculating a loss function of a fault analysis model
Figure 450381DEST_PATH_IMAGE034
Wherein，
Figure 591512DEST_PATH_IMAGE036
Is a first
Figure 89489DEST_PATH_IMAGE037
A fault-like label, T is the number of fault category labels,
Figure 861136DEST_PATH_IMAGE038
is as follows
Figure 811775DEST_PATH_IMAGE037
Probability of class failure.
In summary, according to the method for training the production line fault analysis model, the data set is obtained, the data set comprises the training set and the testing set, the fault analysis model is constructed, the data in the training set is used for training the fault analysis model, so that the loss function is minimum, the trained fault analysis model is tested by using the data in the testing set, the final fault analysis model is obtained, the recognition capability of the fault analysis model on the fault mode corresponding to the fault type can be improved, and the fault type generated on the processing production line can be well predicted.
FIG. 2 is a schematic flow diagram of a method of line fault analysis in accordance with some embodiments of the invention. As shown in fig. 2, the production line fault analysis method includes the following steps:
s201, acquiring data to be analyzed, wherein the data to be analyzed comprises defect data and environment data on a production line;
s202, carrying out fault analysis on the production line according to the data to be analyzed by utilizing the fault analysis model obtained by the training method of the production line fault analysis model.
In summary, according to the production line fault analysis method, the fault analysis model obtained by the training method of the production line fault analysis model can improve the accuracy of fault analysis.
Fig. 3 is a schematic structural diagram of an electronic device according to some embodiments of the invention. As shown in fig. 3, the electronic device 100 includes a memory 110, a processor 120, and a computer program 130 stored in the memory 110, wherein when the computer program 130 is executed by the processor 120, the method for training the line fault analysis model or the method for analyzing the line fault is implemented.
The electronic device according to the embodiment of the present invention may improve the accuracy of the fault analysis when the training method for the production line fault analysis model stored in the memory thereof or the computer program for implementing the production line fault analysis method is executed by the processor.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, the various steps or methods may be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
In the description of the present invention, it is to be understood that the terms "first", "second" and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implying any number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless explicitly specified otherwise.
In the present invention, unless otherwise expressly stated or limited, the first feature "on" or "under" the second feature may be directly contacting the first and second features or indirectly contacting the first and second features through an intermediate. Also, a first feature "on," "over," and "above" a second feature may be directly or indirectly over the second feature or may simply be at a higher level than the second feature. A first feature being "under," "beneath," and "beneath" a second feature may be directly under or obliquely under the second feature, or may simply be at a lesser level than the second feature.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.

Claims (10)

1. A training method for a production line fault analysis model is characterized by comprising the following steps:
acquiring a data set, wherein the data set comprises a training set and a testing set, and the data in the data set comprises defect data detected by automatic optical detection equipment on a production line, environment data of the environment where the production line is located and corresponding fault category labels;
the method comprises the steps of constructing a fault analysis model, wherein the fault analysis model comprises a factorization machine module, a deep neural network module, a feature selection module and a classification module, input data of the factorization machine module and input data of the deep neural network module are both the defect data and the environment data, low-order features and high-order features are correspondingly output, the feature selection module is used for carrying out feature selection on the low-order features and the high-order features, and the classification module is used for obtaining probability distribution of fault categories according to the selected features;
and training the fault analysis model by using the data in the training set to minimize a loss function, and testing the trained fault analysis model by using the data in the testing set to obtain a final fault analysis model.
2. The method of claim 1, wherein the obtaining a data set comprises:
counting defect data, environment data and fault category labels in a plurality of time periods, wherein a defect data set is obtained for each time period
Figure 615217DEST_PATH_IMAGE001
Environment data collection
Figure 10426DEST_PATH_IMAGE002
And a corresponding fault category label, and,
Figure 372268DEST_PATH_IMAGE003
as the number of the j-th type defects,
Figure 460310DEST_PATH_IMAGE004
as to the number of defect classes,
Figure 727343DEST_PATH_IMAGE005
for the i-th type of environment data,
Figure 660664DEST_PATH_IMAGE006
the number of categories of environmental data.
3. A method for training a production line fault analysis model according to claim 2, characterized in that before inputting the data in the data set to the fault analysis model, the statistical data in each time period is preprocessed as follows:
normalizing various types of defect data in the defect data set to obtain a normalized defect data set
Figure 391860DEST_PATH_IMAGE007
Wherein
Figure 650803DEST_PATH_IMAGE008
normalized number of j-th class defects;
carrying out hot independent coding on various types of environmental data in the environmental data set to obtain a hot independent coding set of the environmental data
Figure 139553DEST_PATH_IMAGE009
Wherein
Figure 142144DEST_PATH_IMAGE010
encoding the thermal independent code of the ith type environment data;
carrying out Dense embedding processing on the thermal independent coding set to obtain a low-dimensional Dense representation set
Figure 478579DEST_PATH_IMAGE011
；
Will be provided with
Figure 642844DEST_PATH_IMAGE012
And
Figure 884469DEST_PATH_IMAGE013
splicing to obtain the characteristic vector of the input data
Figure 487489DEST_PATH_IMAGE014
Is marked as
Figure 865380DEST_PATH_IMAGE015
Wherein
Figure 466126DEST_PATH_IMAGE016
is the number of data in X.
4. The method as claimed in claim 3, wherein the input data feature vector of a time segment is selected for training in each training period, and wherein the factorizer module obtains the low-order feature y of the input data feature vector X by the following formula FM ：
Figure 929468DEST_PATH_IMAGE017
，
Wherein,
Figure 83982DEST_PATH_IMAGE018
for the p-th input data feature in X,
Figure 581959DEST_PATH_IMAGE019
and
Figure 88027DEST_PATH_IMAGE020
are parameters of the factorizer module and,
Figure 304245DEST_PATH_IMAGE021
is the product of two k-dimensional vectors.
5. The method as claimed in claim 4, wherein the deep neural network module comprises a plurality of serially connected feedforward neural network modules, each feedforward neural network module comprises a full-link layer, a ReLU activation function and a Dropout layer, and the deep neural network module obtains the higher-order feature y of the input data feature vector X by FNN ：
Figure 249067DEST_PATH_IMAGE022
。
6. The method for training the production line fault analysis model as claimed in claim 5, wherein the feature selection module is a feature selection module of a fusion attention mechanism, and the feature selection module obtains the attention selection feature by the following formula:
Figure 601551DEST_PATH_IMAGE023
，
wherein,
Figure 278520DEST_PATH_IMAGE024
the feature is selected for the purpose of said attention,
Figure 716454DEST_PATH_IMAGE025
，
Figure 215700DEST_PATH_IMAGE026
，
Figure 422690DEST_PATH_IMAGE027
，
Figure 536140DEST_PATH_IMAGE028
in order to splice the feature vectors,
Figure 523687DEST_PATH_IMAGE029
、
Figure 747995DEST_PATH_IMAGE030
and
Figure 75072DEST_PATH_IMAGE031
is a preset weight matrix and is used for carrying out weight adjustment,
Figure 359422DEST_PATH_IMAGE032
the dimension of the vector matrix K.
7. The method for training the production line fault analysis model as claimed in claim 6, wherein the classification module comprises a fully connected layer and a softmax layer, and the classification module obtains the probability distribution of fault classes according to attention selection features by the following formula:
Figure 319419DEST_PATH_IMAGE033
，
where Z is the probability distribution of the fault category.
8. The method for training a production line fault analysis model according to claim 1, wherein the loss function of the fault analysis model is calculated by the following formula
Figure 347418DEST_PATH_IMAGE034
：
Figure 529001DEST_PATH_IMAGE035
，
Wherein,
Figure 46570DEST_PATH_IMAGE036
is as follows
Figure 946393DEST_PATH_IMAGE037
A fault-like label, T is the number of fault category labels,
Figure 512503DEST_PATH_IMAGE038
is a first
Figure 361642DEST_PATH_IMAGE037
Probability of class failure.
9. A production line fault analysis method is characterized by comprising the following steps:
acquiring data to be analyzed, wherein the data to be analyzed comprises defect data and environmental data on a production line;
the fault analysis model obtained by the training method of the production line fault analysis model according to any one of claims 1 to 8 is used for carrying out fault analysis on the production line according to the data to be analyzed.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory, which, when executed by the processor, implements a method of training a line fault analysis model according to any one of claims 1-8, or implements a line fault analysis method according to claim 9.
CN202211060783.3A 2022-08-31 2022-08-31 Training method of production line fault analysis model and production line fault analysis method Pending CN115129019A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211060783.3A CN115129019A (en) 2022-08-31 2022-08-31 Training method of production line fault analysis model and production line fault analysis method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211060783.3A CN115129019A (en) 2022-08-31 2022-08-31 Training method of production line fault analysis model and production line fault analysis method

Publications (1)

Publication Number Publication Date
CN115129019A true CN115129019A (en) 2022-09-30

Family

ID=83387256

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211060783.3A Pending CN115129019A (en) 2022-08-31 2022-08-31 Training method of production line fault analysis model and production line fault analysis method

Country Status (1)

Country Link
CN (1) CN115129019A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117289685A (en) * 2023-11-27 2023-12-26 青岛创新奇智科技集团股份有限公司 Production line fault prediction and self-healing method and system based on artificial intelligence
CN117575565A (en) * 2023-11-15 2024-02-20 重庆雄川科技有限公司 Automatic equipment fault monitoring method and system based on industrial Internet

Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108427986A (en) * 2018-02-26 2018-08-21 中车青岛四方机车车辆股份有限公司 A kind of production line electrical fault prediction technique and device
CN109946606A (en) * 2019-04-03 2019-06-28 四川大学 Vibrating motor defect Fault Classification and device based on convolutional neural networks
CN110320892A (en) * 2019-07-15 2019-10-11 重庆邮电大学 The sewage disposal device fault diagnosis system and method returned based on Lasso
US20210073973A1 (en) * 2019-09-06 2021-03-11 Beijing Baidu Netcom Science Technology Co., Ltd. Method and apparatus for component fault detection based on image
CN112907562A (en) * 2021-03-16 2021-06-04 中科海拓(无锡)科技有限公司 MobileNet-based SMT defect classification algorithm
CN113591948A (en) * 2021-07-16 2021-11-02 清华大学 Defect pattern recognition method and device, electronic equipment and storage medium
CN113740776A (en) * 2021-08-17 2021-12-03 广州番禺电缆集团有限公司 Cable sheath grounding circulation fault prediction method, device, equipment and storage medium
CN113837263A (en) * 2021-09-18 2021-12-24 浙江理工大学 Gesture image classification method based on feature fusion attention module and feature selection
CN113935403A (en) * 2021-09-24 2022-01-14 华中科技大学 Multi-label multi-classification based main shaft bearing and cutter composite fault diagnosis method
CN114819102A (en) * 2022-05-19 2022-07-29 东南大学溧阳研究院 GRU-based air conditioning equipment fault diagnosis method

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108427986A (en) * 2018-02-26 2018-08-21 中车青岛四方机车车辆股份有限公司 A kind of production line electrical fault prediction technique and device
CN109946606A (en) * 2019-04-03 2019-06-28 四川大学 Vibrating motor defect Fault Classification and device based on convolutional neural networks
CN110320892A (en) * 2019-07-15 2019-10-11 重庆邮电大学 The sewage disposal device fault diagnosis system and method returned based on Lasso
US20210073973A1 (en) * 2019-09-06 2021-03-11 Beijing Baidu Netcom Science Technology Co., Ltd. Method and apparatus for component fault detection based on image
CN112907562A (en) * 2021-03-16 2021-06-04 中科海拓(无锡)科技有限公司 MobileNet-based SMT defect classification algorithm
CN113591948A (en) * 2021-07-16 2021-11-02 清华大学 Defect pattern recognition method and device, electronic equipment and storage medium
CN113740776A (en) * 2021-08-17 2021-12-03 广州番禺电缆集团有限公司 Cable sheath grounding circulation fault prediction method, device, equipment and storage medium
CN113837263A (en) * 2021-09-18 2021-12-24 浙江理工大学 Gesture image classification method based on feature fusion attention module and feature selection
CN113935403A (en) * 2021-09-24 2022-01-14 华中科技大学 Multi-label multi-classification based main shaft bearing and cutter composite fault diagnosis method
CN114819102A (en) * 2022-05-19 2022-07-29 东南大学溧阳研究院 GRU-based air conditioning equipment fault diagnosis method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
刘树春等: "《深度实践OCR基于深度学习的文字识别》", 31 May 2020 *
纪宇晨: "基于机器学习的无线局域网入侵检测模型研究", 《中国优秀硕士学位论文全文数据库信息科技辑（月刊）》 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117575565A (en) * 2023-11-15 2024-02-20 重庆雄川科技有限公司 Automatic equipment fault monitoring method and system based on industrial Internet
CN117289685A (en) * 2023-11-27 2023-12-26 青岛创新奇智科技集团股份有限公司 Production line fault prediction and self-healing method and system based on artificial intelligence
CN117289685B (en) * 2023-11-27 2024-02-02 青岛创新奇智科技集团股份有限公司 Production line fault prediction and self-healing method and system based on artificial intelligence

Similar Documents

Publication Publication Date Title
CN110415215B (en) Intelligent detection method based on graph neural network
CN111552609B (en) Abnormal state detection method, system, storage medium, program and server
CN111382542B (en) Highway electromechanical device life prediction system facing full life cycle
CN115129019A (en) Training method of production line fault analysis model and production line fault analysis method
CN113702895B (en) Online quantitative evaluation method for error state of voltage transformer
CN115308562A (en) Chip testing method and related equipment
CN113837000A (en) Small sample fault diagnosis method based on task sequencing meta-learning
Khuat et al. Ensemble learning for software fault prediction problem with imbalanced data.
CN112767106B (en) Automatic auditing method, system, computer readable storage medium and auditing equipment
CN117726240B (en) Quality evaluation classification method and system based on convolutional neural network
CN112763215B (en) Multi-working-condition online fault diagnosis method based on modular federal deep learning
CN114139589A (en) Fault diagnosis method, device, equipment and computer readable storage medium
CN116611745A (en) Plastic film quality evaluation method and system based on intelligent manufacturing
CN115757103A (en) Neural network test case generation method based on tree structure
CN117743909A (en) Heating system fault analysis method and device based on artificial intelligence
CN113344915B (en) Method and system for detecting defects of key parts of power transmission line
CN115687330A (en) Automobile fault code analysis method, storage medium and electronic equipment
CN111160419B (en) Deep learning-based electronic transformer data classification prediction method and device
WO2021126591A1 (en) Method for physical system anomaly detection
CN114692887A (en) Semi-supervised learning system and semi-supervised learning method
CN111881040A (en) Test data generation method of abstract state model based on recurrent neural network
Wang et al. Classification and prediction of wafer probe yield in DRAM manufacturing using Mahalanobis-Taguchi system and neural network
Oh et al. LSTM-based PdM Platform for Automobile SCU Inspection Equipment
CN118277941B (en) Intelligent discovery method for abnormal price of civil aviation passenger ticket based on self-growth mechanism
CN118211171B (en) Knowledge graph-based target path mining method

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
RJ01 Rejection of invention patent application after publication

Application publication date: 20220930

RJ01 Rejection of invention patent application after publication