CN113759838A - Method and device for predicting shot blasting quality - Google Patents
Method and device for predicting shot blasting quality Download PDFInfo
- Publication number
- CN113759838A CN113759838A CN202011213347.6A CN202011213347A CN113759838A CN 113759838 A CN113759838 A CN 113759838A CN 202011213347 A CN202011213347 A CN 202011213347A CN 113759838 A CN113759838 A CN 113759838A
- Authority
- CN
- China
- Prior art keywords
- data
- training
- shot blasting
- forest
- real
- 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
- 238000000034 method Methods 0.000 title claims abstract description 100
- 238000005422 blasting Methods 0.000 title claims abstract description 77
- 238000012549 training Methods 0.000 claims abstract description 73
- 230000008569 process Effects 0.000 claims abstract description 66
- 238000012545 processing Methods 0.000 claims description 26
- 238000003066 decision tree Methods 0.000 claims description 20
- 238000012360 testing method Methods 0.000 claims description 20
- 238000007637 random forest analysis Methods 0.000 claims description 18
- 238000004422 calculation algorithm Methods 0.000 claims description 11
- 238000002790 cross-validation Methods 0.000 claims description 11
- 238000005480 shot peening Methods 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000000605 extraction Methods 0.000 claims description 7
- 238000009499 grossing Methods 0.000 claims description 6
- 238000012544 monitoring process Methods 0.000 claims description 5
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000013500 data storage Methods 0.000 claims description 4
- 238000003745 diagnosis Methods 0.000 claims description 4
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 description 11
- 238000004519 manufacturing process Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 238000004891 communication Methods 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 239000000463 material Substances 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000003860 storage Methods 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000001514 detection method Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000002068 genetic effect Effects 0.000 description 2
- 238000002347 injection Methods 0.000 description 2
- 239000007924 injection Substances 0.000 description 2
- 230000007246 mechanism Effects 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 238000003908 quality control method Methods 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 238000005728 strengthening Methods 0.000 description 2
- 239000013598 vector Substances 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000005482 strain hardening Methods 0.000 description 1
- 230000003746 surface roughness Effects 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
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
- G05B2219/32—Operator till task planning
- G05B2219/32368—Quality control
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/02—Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]
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 method and a device for predicting shot blasting quality, wherein the method comprises the following steps: the method comprises the steps of collecting real-time data of the shot blasting equipment, positioning the real-time data to obtain controllable parameters, inputting the controllable parameters into a prediction model, and outputting results, wherein the prediction model is obtained by training with a deep forest, and finally, the controllable parameters are controlled in real time, so that the output of the prediction result after the training model is applied can be ensured to meet the requirements of process specification parameters. The invention relates to the technical field of shot blasting, and compared with the prior art, the invention has the advantages that: easy to realize and saves cost.
Description
Technical Field
The invention relates to the technical field of shot blasting, in particular to a method and a device for predicting shot blasting quality.
Background
Shot blasting is a surface strengthening process widely adopted in factories, particularly in industries with high requirements on surface strength, such as airplane manufacturing, engineering machinery and the like, is a cold working process for bombarding the surface of a workpiece by using shot particles and implanting residual compressive stress to improve the fatigue strength of the workpiece, and is widely used for improving the mechanical strength, wear resistance, fatigue resistance, corrosion resistance and the like of parts. The shot blasting process has important significance in high and new technologies such as aerospace, engineering machinery and the like. However, for a complex process with interdependence of the sub-forming process and interdependence of multiple process parameters, how to clearly understand the association rule between the process parameters is one of the key problems of the quality control of the forming process. Therefore, the method has important significance for improving the shot blasting quality by digging the process control rules hidden in the shot blasting process, optimizing the process parameters and predicting the forming quality.
The existing method for testing shot blasting quality and adjusting shot blasting unification is to operate and adjust through an almen test piece, fix a test piece clamp on a simulation piece of each shot blasting part specified by a pattern respectively, fix a test piece on the clamp, place the simulation piece on a part tool in a shot blasting chamber, start an operation mechanism of the tool and perform shot blasting. The arc height value of the unloaded test piece is measured by taking the non-shot blasting surface as a reference surface, and the unloaded shot blasting test piece cannot be reused. New coupons were reinstalled, shot blasted and arc height measured. After shot blasting for different times (or shot blasting times) on N test pieces, an arc height curve is obtained, and the shot blasting intensity is determined according to the curve. The adjustment of the shot blasting intensity mainly defines some core process parameters (such as shot flow and the like), and meanwhile, the preset standard shot blasting process parameters are used for operation, so whether the expected shot blasting intensity can be achieved or not is a key core problem of quality control in the field of shot blasting.
Therefore, the method for judging whether the shot blasting quality reaches the standard mainly by a post detection means has the defects of difficult repair, resource waste of enterprises and irreparable loss of the enterprises if the quality is in problem.
Patent publication No. CN109508488A discloses a shot peening forming process parameter prediction method based on genetic algorithm optimization BP neural network, which is used for solving the technical problem of poor practicability of the existing shot peening forming method. The technical scheme is that a BP neural network is adopted to establish a complex nonlinear mapping relation among part appearance characteristics, material mechanical properties and shot blasting process parameters, and then a genetic algorithm is adopted to optimize the structure and the parameters of the BP neural network. The scheme mainly adopts the relation between the part material and the shot blasting process to predict the shot blasting process parameters.
Patent publication No. CN111401623A discloses a shot peening surface integrity prediction method based on BP neural network, which optimizes the previous technical scheme.
However, the above two methods for predicting the shot blasting quality by using the BP neural network all face the following problems:
(1) the model interpretability determines the final floor application of the model to a certain extent in the industrial scene application process, and the BP deep neural network is a black box in a strict sense and is difficult to interpret.
(2) The neural network model training stage needs a large number of sample sets, and the acquisition of positive and negative samples in an industrial scene is very challenging and needs to consume large resources.
(3) From the application deployment of the model, the BP neural network algorithm is not suitable for parallel computing, cannot utilize hadoop cluster resources, and can only be accelerated by the GPU, which is also very cost-consuming.
Disclosure of Invention
The invention aims to overcome the technical defects and provide a method for predicting the shot blasting quality, which is easy to realize and saves cost.
In addition, the application also provides a shot blasting quality prediction device so as to ensure the realization and the application of the method in practice.
In a first aspect of the present application, there is provided a method of predicting peen quality, the method comprising the steps of:
collecting real-time data of shot blasting equipment, wherein the real-time data is real-time working condition data of the shot blasting equipment in the working process;
positioning the real-time data to obtain controllable parameters, inputting the controllable parameters into a prediction model, and outputting a result, wherein the prediction model is obtained by training through the following steps:
step 1: constructing a shot blasting quality prediction sample data set, wherein the shot blasting quality prediction sample data set comprises training characteristics extracted by preprocessing process specification parameters and historical real-time working condition data and extracting the characteristics;
step 2: dividing the shot blasting quality prediction sample data set to obtain a training set and a test set;
and step 3: inputting the training set into a deep forest for training, wherein the deep forest is composed of cascaded forests and multi-granularity sliding windows,
the multi-granularity sliding window obtains a sliding sample example through scanning a training set, the cascade forest is composed of a plurality of levels of forests, the cascade forest receives the sliding sample example and outputs a corresponding predicted value, and the mean value of regression results of decision trees of the last level of forest is the prediction result of the sliding sample example;
and the controllable parameters are controlled in real time, so that the output prediction result after the training model is applied meets the requirement of the technological specification parameters.
Preferably, after each level of forest is trained, the cascade forest performs 10-fold cross validation on the training set, records 10-fold cross validation results, and continues to generate the next level of forest, and when the deviation of two adjacent 10-fold cross validation results is smaller than an expected value, the training is stopped.
Preferably, each level of the cascade of forests comprises a random forest and a fully random tree forest.
Preferably, the complete random forest is composed of a plurality of complete decision trees, each node in the complete decision trees randomly selects a feature as a splitting feature, and continuously grows the whole tree until all the remaining sliding sample instances belong to the same class or the number of samples is less than 10;
the random forest is composed of a plurality of random trees, each node in the random trees randomly selects sqrt (d) features, wherein d is the total number of input features, the features of MSE are selected as splitting features of the node, and the whole tree is grown at the same time until all the remaining samples belong to the same class or the number of the samples is less than 10.
Preferably, the preprocessing includes truncation processing and smoothing processing, wherein the truncation processing is to truncate a signal segment of the real-time working condition data when the shot blasting is stable, and the smoothing processing includes missing value filling and using a filtering algorithm to remove a high-frequency signal.
Preferably, the feature extracted features include time domain features, frequency domain features, and time-frequency domain analysis.
In a second aspect of the present application, there is provided an apparatus for predicting shot quality, comprising:
the data acquisition module is used for continuously acquiring real-time working condition data in the operation of the shot blasting equipment and receiving process specification parameters;
the training module is used for obtaining a prediction model through deep forest training of the training set and testing and optimizing the prediction model through the test set;
the data processing module is used for carrying out data conversion of feature extraction on the real-time working condition data set process specification parameters;
and the data storage module is used for storing the data acquired by the data acquisition module, so as to be used as a basis for the data processing module and the training module, and simultaneously storing the output result generated by the training module.
Preferably, the system further comprises a field control module, which is used for monitoring the real-time data acquired by the data acquisition module in real time and carrying out real-time diagnosis reminding according to the prediction result of the training module.
Preferably, the system further comprises an edge calculation module, wherein the edge calculation module is used for loading the real-time working condition data of the shot blasting equipment into the prediction model to perform edge calculation;
and the network application module is used for transmitting data among the modules.
Compared with the prior art, the process has the advantages that: the method includes the steps of collecting process parameters in a shot blasting process at a high speed in real time through the Internet of things, obtaining a mechanism relation among input process parameters, extracting basic features in data based on a machine learning data processing technology, training the data, and reversely generating a shot blasting quality prediction model.
(1) Because the model training is carried out by adopting the integrated learning modes of random forests, deep forests and the like, the random forests can provide different interpretations of the decision tree, and compared with a neural network, the variable amplification learning mode can be used for amplifying variables for learning, the characteristics are easy to distinguish, so that the variable amplification learning mode can be better interpreted and can be practically applied;
(2) the method can be applied to a small sample set scene, and good performance can be ensured;
(3) the random forest computing cost is low, and training is completed without depending on a GPU (graphics processing unit), so that the computing speed is higher in the actual application process, and the operation supervision of the shot blasting equipment can be better adapted. By the method, a shot blasting process optimization model and a quality prediction model can be constructed in advance, and the problems of quality found by post inspection and the waste of reworking and repairing personnel, working hours and materials are solved.
Drawings
FIG. 1 is a flow chart of a method of predicting shot quality in accordance with the present invention.
FIG. 2 is a diagram of a process specification parameter acquisition for predicting shot quality in accordance with the present invention.
FIG. 3 is a schematic diagram of the present invention for extracting the real-time behavior data feature for predicting shot quality.
FIG. 4 is a diagram of a cascaded forest structure for predicting shot blasting quality according to the present invention.
FIG. 5 is a schematic flow chart of a decision tree for predicting shot quality according to the present invention.
FIG. 6 is a block diagram showing the construction of an apparatus for predicting shot quality according to the present invention.
Detailed Description
The following description is only a preferred embodiment of the present invention, and does not limit the scope of the invention, which is further described with reference to the accompanying drawings and embodiments.
For the understanding of the embodiment, the method for predicting the shot blasting quality based on the random forest algorithm prediction model adopted by the invention in the actual processing process is described in detail below.
Example 1:
as shown in fig. 1, a flowchart of a method for predicting shot blasting quality according to an embodiment of the present invention is specifically as follows:
(1) collecting real-time data of shot blasting equipment, wherein the real-time data is real-time working condition data of the shot blasting equipment in the working process; specifically, the real-time working condition data is a process parameter of the shot blasting equipment in the working process, and the process parameter collected in the embodiment includes a jet pressure, a distance from the surface to the nozzle, a jet angle, a shot flow rate, a shot blasting moving speed and a gas flow rate, as shown in table 1:
TABLE 1
The process parameters are all referenced by time series, so the historical data is also time series data of the real-time data, and the logical relationship among the process parameters can be obtained through better comparison of the time series data.
(2) Positioning the real-time data to obtain controllable parameters, inputting the controllable parameters into a prediction model, and outputting a result; specifically, the controllable parameters are shot flow, distance between the workpiece and the nozzle, injection pressure, shot blasting time and the like, the controllable parameters serve as an input set of a prediction model, and an output set of the prediction model is a residual stress value, surface roughness, surface coverage and the like.
Next, the present embodiment takes the residual stress value as an example, and specifically describes how to train the prediction model by using the deep forest, where the training of the prediction model includes the following steps:
step 1: constructing a shot blasting quality prediction sample data set, wherein the shot blasting quality prediction sample data set comprises training characteristics extracted by preprocessing process specification parameters and historical real-time working condition data and extracting the characteristics;
therefore, two parts of the shot blasting quality prediction sample data set are formed, namely, the process specification parameters in the shot blasting process, the real-time working condition data acquired in the shot blasting process, the characteristic values after preprocessing and characteristic extraction are shown in the 240 samples in the figures 2 and 3, and each sample has 195-dimensional characteristics.
The process of preprocessing the real-time working condition data comprises the following steps: and performing truncation processing and smoothing processing, wherein the truncation processing is to truncate a signal section of the real-time working condition data when the shot blasting is stable, and the smoothing processing comprises missing value filling and filtering algorithm removal of high-frequency signals. And the preprocessing of the data may also include problem processing such as data missing, noise, consistency, redundancy, data equalization, outliers, data duplication and the like.
The features included in the feature extraction are time domain features, frequency domain features and time-frequency domain analysis, the three features are developed from three dimensions of the digital signal, and the range of the features is basically covered.
Specifically, the time domain signature envelope: mean, root mean square value, variance, kurtosis, peak, impulse, skewness, peak-to-peak value, coefficient of variation, and absolute median potential difference;
the frequency domain features include: frequency, amplitude, phase, power spectrum, energy spectrum, cepstrum, etc.;
the time-frequency domain analysis comprises: wavelet coefficients, energy spectra, IMF, etc.
Step 2: after the feature extraction is completed, the shot blasting quality prediction sample data set is divided into 3: 1, dividing the test set into a training set and a test set in proportion, wherein the residual stress prediction training set comprises 180 cases and the test set comprises 60 cases after division;
and step 3: inputting the training set into a deep forest for training, and detailing a deep forest model training process by combining a shot blasting process.
Referring to fig. 4, the deep forest algorithm is composed of two key components, namely a cascade forest component and a multi-granularity sliding window component, and after the original features are scanned in a multi-granularity mode through the multi-granularity sliding window component, enhanced feature vectors including multiple scales of local structures of the features can be obtained. And each level of cascade forest uses a group of forest to simulate characteristics, the forest of each level uses the information of the previous level as the input information of the forest, and the output of the forest of each level provides the input information for the next level.
In the invention, on the basis of original features, the features are extracted by using multi-granularity sliding windows with different sizes, and for the feature of a training set with the dimension of 195, if the training set is scanned by using a sliding window with the size of 100, 96 sliding sample examples can be obtained, wherein the dimension of each example is 100; if a sliding window of size 50 is used, 146 sample instances can be obtained, each instance having a dimension of 50; similarly, assuming a window size of 50, a feature of 195 dimensions will yield 146 new instances of the feature. For 1D vector features, N ═ D-W + 1; n is the sliding sample example, D is the original feature dimension, and W is the window size.
Taking a 100-dimensional window as an example, 96 sliding sample examples are obtained, each sliding sample example represents a local structure or a sub-sampling sample, a training set has 180 samples, each sliding sample example can obtain 180 training samples, the 180 100-dimensional features are used as the input of a random forest, and a prediction result can be obtained and input into the next-level random forest.
In the cascade forest network, after the training of the primary network is finished, the training data is subjected to 10-fold cross validation once, and after the training of the primary network, the corresponding prediction type can be output, the 10-fold cross validation with the real type is performed, and the 10-fold cross validation result is recorded. And continuing to generate the next level, similarly performing cross validation on the result of the level, if the result of the cross validation of the level is smaller than the expected value compared with the result of the previous level, stopping generating a new level by the cascade forest, and terminating training, wherein the expected value is determined by working personnel according to working conditions. Thus in a cascaded forest structure its level can be determined automatically.
The deep forest can better cope with a small sample set and a multi-granularity sliding window mode, the function of the deep forest is similar to that of one-time upsampling operation, one sample is changed into N sliding example samples, and the samples represent the local structure of an original sample to a certain extent, so that the multi-granularity sliding window operation can be called as structured upsampling, and the representation capability of a model is enhanced.
A cascaded forest structure is shown in fig. 4, each level comprising random forests and fully random tree forests, each forest outputting the final predicted residual stress and then concatenating it to re-represent the original input if it is desired to predict the residual stress of the surface of the peened workpiece.
Specifically, the fully random forest is composed of 100 complete decision trees, each node in the tree randomly selects a feature as a splitting feature, and the whole tree is continuously grown until all remaining samples belong to the same class or the number of samples is less than 10. The random forest is composed of 100 random trees, each node in the tree randomly selects sqrt (d) features, wherein d is the total number of input features, then selects the feature with the best MSE as the splitting feature of the node, and grows the whole tree until all the remaining samples belong to the same class or the number of the samples is less than 10.
Finally, each forest will calculate the prediction value of the training sample whose associated instance falls into the leaf node, and then calculate the average value of all decision trees in the forest to obtain the prediction of the shot peening residual stress.
In the training process, a large number of decision tree branches are generated, wherein the decision tree branches (i.e. the training process of the decision tree) take the MSE index of the data set as a reference, and the computation formula of MSE is shown in the following figure, where n is the number of samples, m is the number of features, and x is the number of featuresijIs the value of the j-th feature of the i-th sample, ujIs the average of the jth sample. From the above equation, for any data set, there is a mean square error value greater than zero, and when the eigenvalues of all samples are equal to the mean of the eigenvalues, the mean square error is the smallest (i.e., 0).
The decision tree branching aims at branching a complete data set for a limited number of times according to the value range of each characteristic to finally obtain a plurality of subsets, so that the MSE value of each subset is as small as possible, or the number of samples in the subsets is small enough, and the condition for stopping the decision tree branching is as follows:
1. when the MSE value of the grouped data set is 0, the data set is not branched;
2. when the number of data set samples after branching is less than a certain threshold (hyper-parameter), the data set is no longer branched.
Taking fig. 5 as an example, the number of original total samples is 96, MSE is 256, and the mean value of residual stress is-391. And performing MSE calculation after branching according to each characteristic and value range to obtain the maximum descending amplitude of the branched MSE when the value of the average flow characteristic is about 0.77. Therefore, branching is performed based on this, and samples with an average flow rate of 0.77 or more among all samples are branched to the left (54 samples shown in fig. 5), and the remaining samples are branched to the right (42 samples shown in fig. 5).
The above process is repeated for the two subsets after branching, and when the MSE value of the subset is 0, or the number of samples in the subset is less than 10, the branching stops. The process of decision tree branching is the training process of the decision tree model.
During prediction of the decision tree model, the sample to be predicted is distributed to the sub-set (leaf node) according to the characteristics of the sample to be predicted, and the residual stress mean value of the sub-set is the predicted value of the residual stress of the sample. Taking the decision tree of the above graph as an example, a sample with a shot flow FFT variance of 0.8 and a median injection pressure of 3 would branch to the left in the first branch due to the average flow rate being greater than 0.77, and would branch to the left in the second branch due to the average pressure being greater than 3, and would be finally assigned to the leftmost subset (leaf node), so its predicted residual stress mean value is-244.
In this embodiment, the deep forest overall model training process specifically includes: firstly, 195-dimensional features of different samples are scanned according to a multi-granularity sliding window, such as a 100-dimensional window, so as to respectively obtain 96 sliding sample examples, wherein each feature example is 100-dimensional. Similarly, assuming a window size of 50, a feature of 195 dimensions will yield 146 new instances of the feature.
After 96 100-dimensional feature examples are obtained, through two fully random forests A and fully random forest B, each forest outputs a corresponding predicted value for each input sample, so that 96 predicted values are output by 96 feature examples, and the predicted values of the feature examples generated by the two random forests are cascaded to obtain 192 features with the feature number of (96 × 1 × 2).
Because the input is an original 195-dimensional feature of a sample, by setting three sliding windows with different sizes, 96 feature examples (the dimension is 100) are obtained by the sliding window with 100 dimensions, 146 feature examples are obtained by the sliding window with 50 dimensions, 116 feature examples are obtained by the sliding window with 80 dimensions, each feature example is mapped into a single predicted value after passing through a forest, the feature dimension of the 96 feature examples is 192 after passing through 2 forest, the feature dimension of the 146 feature examples is 292 after passing through 2 forest, the feature dimension of the 116 feature examples is 232 after passing through 2 forest, all features are finally integrated, and the total feature number obtained by each sample is as follows: 192+292+232, 716.
Finally, 716 dimensions are used as input of the forest waterfall, and the input of each layer of forest except the first layer of forest is as follows: the output result (4) + original features (716) of the forest is 720 features. And taking the mean value of the regression results of all decision trees of the last layer of forest as the prediction result of the sample.
In order to reduce the risk of overfitting, each forest prediction output is generated by 10-fold, each sample is used as training data for 9 times, 9 times of prediction values are generated, and the prediction values are averaged to be used as the input of the enhancement features of the next stage in the cascade. After a new stage is extended, the performance of the entire cascade is estimated on the validation set, and if there is no significant performance gain, the training process is terminated; thus, the number of stages in the cascade is automatically determined.
From the training process described above, it is known that in contrast to most deep neural networks where the complexity of the model is fixed, deep forests can properly determine the complexity of their model by terminating the training. This enables the deep forest to be adapted to different scales of training data, without being limited to large scale training data.
Finally, referring to table 2, it can be seen that the accuracy of the deep forest is greater than that of the BP neural network intuitively in the form of R2.
Table 2:
| serial number | | R2 formula | |
| 1 | gcforest | 89.50％ | |
| 2 | forest | 85.70％ | |
| 3 | BP neural network | 86.40％ |
Wherein the formula for the formula R2 is shown below, yiTo test the residual stress values of the ith sample of the set,
the residual stress predicted value calculated by the model according to the ith sample characteristic input is used,
the average value of the residual stress of the sample is the test set. Closer to 100% this value indicates higher accuracy.
(3) And the controllable parameters are controlled in real time, so that the output prediction result after the training model is applied meets the requirement of the technological specification parameters.
In specific production, a standard AI quality judgment algorithm model can be constructed based on the relation direction of process parameters in a prediction model, the model is applied to real-time control of a field production process, IOT data of shot blasting equipment is monitored based on a logic relation, and quality risk problems in the shot blasting process are warned in time according to input and output parameter changes.
Specifically, training can be performed based on IOT original process parameters acquired on site, and a training set is constructed; extracting the IOT process parameter characteristics based on a standard algorithm to form a basic model; continuously training, strengthening and optimizing the model to form a quality diagnosis and process optimization model with an application scene; and finally, loading the model and configuring a field HMI (human machine interface), monitoring the process parameters in real time, and diagnosing and prompting the production quality in real time.
Example 2:
corresponding to the method, the embodiment of the application also provides a corresponding device for realizing the method. The device is explained below with reference to fig. 6.
Referring to fig. 6, fig. 6 is a view showing an apparatus for predicting shot quality according to the present application, the apparatus including: the device comprises a detection data acquisition module, a data storage module, a data processing module and a training module; the function and connection relationship of each module in the device are explained according to the working principle of the device.
The data acquisition module S601 is used for continuously acquiring real-time working condition data in the operation of the shot blasting equipment and receiving process specification parameters, and can be a special internet of things gateway for the shot blasting equipment;
and the training module S604 is used for obtaining a prediction model by deep forest training of the training set, testing and optimizing the prediction model by the test set, and can be an AI model operation server.
And the data processing module S603 is used for performing data conversion of feature extraction on the real-time working condition data set process specification parameters.
The data storage module S602 is used for storing the data acquired by the data acquisition module, so as to be used as a basis for the data processing module and the training module, and simultaneously storing an output result generated by the training module;
in the application process, the device may further include:
and the field control module S605 is used for monitoring the real-time data acquired by the data acquisition module in real time and diagnosing and reminding in real time according to the prediction result of the training module, can be a workshop field HMI, and can realize real-time monitoring of process parameters and real-time diagnosis and reminding of production quality by configuring the workshop field HMI.
In the application process, the device may further include:
and the edge calculation module S606 is used for performing edge operation on the real-time process parameters of the shot blasting equipment IOT at a high speed and timely early warning quality risk problems in the shot blasting process.
And the network application module S607 is used for transmitting data among the modules, is convenient for workers to call, and can be a WEB application server and a plurality of switches.
In the embodiments provided in the present invention, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. The above-described apparatus embodiments are merely illustrative, and for example, the division of the modules is merely a logical division, and there may be other divisions in actual implementation, and for example, a plurality of modules or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection of devices or modules through some communication interfaces, and may be in an electrical, mechanical or other form.
The modules described as separate parts may or may not be physically separate, and parts displayed as modules may or may not be physical modules, may be located in one place, or may be distributed on a plurality of network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment.
In addition, functional modules in the embodiments of the present invention may be integrated into one processing module, or each of the modules may exist alone physically, or two or more modules are integrated into one module. The functions, if implemented in the form of software functional modules and sold or used as a stand-alone product, may be stored in a non-volatile computer-readable storage medium executable by a processor. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-only memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The invention and its embodiments have been described above, without limitation, and what is shown in the drawings is only one of the embodiments of the invention, to which the actual structure is not limited. In summary, those skilled in the art should appreciate that they can readily use the disclosed conception and specific embodiments as a basis for designing or modifying other structures for carrying out the same purposes of the present invention without departing from the spirit and scope of the invention as defined by the appended claims.
Claims (9)
1. A method for predicting shot blasting quality, comprising the steps of:
collecting real-time data of shot blasting equipment, wherein the real-time data is real-time working condition data of the shot blasting equipment in the working process;
positioning the real-time data to obtain controllable parameters, inputting the controllable parameters into a prediction model, and outputting a result, wherein the prediction model is obtained by training through the following steps:
step 1: constructing a shot blasting quality prediction sample data set, wherein the shot blasting quality prediction sample data set comprises training characteristics extracted by preprocessing process specification parameters and historical real-time working condition data and extracting the characteristics;
step 2: dividing the shot blasting quality prediction sample data set to obtain a training set and a test set;
and step 3: inputting the training set into a deep forest for training, wherein the deep forest is composed of cascaded forests and multi-granularity sliding windows,
the multi-granularity sliding window obtains a sliding sample example through scanning a training set, the cascade forest is composed of a plurality of levels of forests, the cascade forest receives the sliding sample example and outputs a corresponding predicted value, and the mean value of regression results of decision trees of the last level of forest is the prediction result of the sliding sample example;
and the controllable parameters are controlled in real time, so that the output prediction result after the training model is applied meets the requirement of the technological specification parameters.
2. A method of predicting shot peening quality as claimed in claim 1, wherein: and after each level of forest is trained, performing 10-fold cross validation on the training set, recording 10-fold cross validation results, continuously generating the next level of forest, and stopping training when the deviation of the two adjacent 10-fold cross validation results is smaller than an expected value.
3. A method of predicting shot peening quality as claimed in claim 2, wherein: each level of the cascaded forest includes a random forest and a fully random tree forest.
4. A method of predicting shot peening quality as claimed in claim 3, wherein: the complete random forest is composed of a plurality of complete decision trees, each node in the complete decision trees randomly selects a feature as a splitting feature, and the whole tree is continuously grown until all the remaining sliding sample instances belong to the same class or the number of samples is less than 10;
the random forest is composed of a plurality of random trees, each node in the random trees randomly selects sqrt (d) features, wherein d is the total number of input features, the features of MSE are selected as splitting features of the node, and the whole tree is grown at the same time until all the remaining samples belong to the same class or the number of the samples is less than 10.
5. A method of predicting shot peening quality as claimed in claim 1, wherein: the preprocessing comprises truncation processing and smoothing processing, wherein the truncation processing is to intercept a signal section of the real-time working condition data when the shot blasting is stable, and the smoothing processing comprises missing value filling and high-frequency signal removal by using a filtering algorithm.
6. A method of predicting shot peening quality as claimed in claim 1, wherein: the feature extraction includes time domain feature, frequency domain feature and time-frequency domain analysis.
7. A shot quality prediction apparatus based on a method of predicting shot quality, characterized in that: the system comprises a data acquisition module, a data processing module and a data processing module, wherein the data acquisition module is used for continuously acquiring real-time working condition data in the operation of shot blasting equipment and receiving process specification parameters;
the training module is used for obtaining a prediction model through deep forest training of the training set and testing and optimizing the prediction model through the test set;
the data processing module is used for carrying out data conversion of feature extraction on the real-time working condition data set process specification parameters;
and the data storage module is used for storing the data acquired by the data acquisition module, so as to be used as a basis for the data processing module and the training module, and simultaneously storing the output result generated by the training module.
8. An apparatus for predicting shot blasting quality as set forth in claim 7, wherein: also comprises
And the field control module is used for monitoring the real-time data acquired by the data acquisition module in real time and carrying out real-time diagnosis reminding according to the prediction result of the training module.
9. An apparatus for predicting shot blasting quality as set forth in claim 8, wherein: also comprises
The edge calculation module is used for loading the real-time working condition data of the shot blasting equipment into the prediction model for edge calculation;
and the network application module is used for transmitting data among the modules.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011213347.6A CN113759838A (en) | 2020-11-04 | 2020-11-04 | Method and device for predicting shot blasting quality |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011213347.6A CN113759838A (en) | 2020-11-04 | 2020-11-04 | Method and device for predicting shot blasting quality |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113759838A true CN113759838A (en) | 2021-12-07 |
Family
ID=78785968
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011213347.6A Pending CN113759838A (en) | 2020-11-04 | 2020-11-04 | Method and device for predicting shot blasting quality |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113759838A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114997549A (en) * | 2022-08-08 | 2022-09-02 | 阿里巴巴（中国）有限公司 | Interpretation method, device and equipment of black box model |
| CN115146529A (en) * | 2022-06-13 | 2022-10-04 | 武汉理工大学 | Method for predicting cold shot blasting strengthening residual stress of automotive suspension spring |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110297469A (en) * | 2019-05-17 | 2019-10-01 | 同济大学 | The production line fault judgment method of Ensemble feature selection algorithm based on resampling |
| CN110728391A (en) * | 2018-07-17 | 2020-01-24 | 广西大学 | Depth regression forest short-term load prediction method based on expandable information |
| CN111401623A (en) * | 2020-03-12 | 2020-07-10 | 重庆大学 | Shot peening strengthening surface integrity prediction method based on BP neural network |
-
2020
- 2020-11-04 CN CN202011213347.6A patent/CN113759838A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728391A (en) * | 2018-07-17 | 2020-01-24 | 广西大学 | Depth regression forest short-term load prediction method based on expandable information |
| CN110297469A (en) * | 2019-05-17 | 2019-10-01 | 同济大学 | The production line fault judgment method of Ensemble feature selection algorithm based on resampling |
| CN111401623A (en) * | 2020-03-12 | 2020-07-10 | 重庆大学 | Shot peening strengthening surface integrity prediction method based on BP neural network |
Non-Patent Citations (1)
| Title |
|---|
| 陈吕鹏等: "基于深度森林算法的电力系统短期负荷预测", 《电力建设》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115146529A (en) * | 2022-06-13 | 2022-10-04 | 武汉理工大学 | Method for predicting cold shot blasting strengthening residual stress of automotive suspension spring |
| CN115146529B (en) * | 2022-06-13 | 2024-04-05 | 武汉理工大学 | Method for predicting cold shot blasting reinforcement residual stress of automobile suspension spring |
| CN114997549A (en) * | 2022-08-08 | 2022-09-02 | 阿里巴巴（中国）有限公司 | Interpretation method, device and equipment of black box model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111323220B (en) | Fault diagnosis method and system for gearbox of wind driven generator | |
| CN107462785B (en) | The more disturbing signal classifying identification methods of power quality based on GA-SVM | |
| CN104537415B (en) | A kind of non-linear process industry failure predication and recognition methodss based on compressed sensing and DROS ELM | |
| CN113759838A (en) | Method and device for predicting shot blasting quality | |
| CN111695209A (en) | Rotary machine small sample health assessment method driven by meta-depth learning | |
| Duan et al. | MS-SSPCANet: A powerful deep learning framework for tool wear prediction | |
| Nouhi et al. | Prediction of surface roughness of various machining processes by a hybrid algorithm including time series analysis, wavelet transform and multi view embedding | |
| CN109409271B (en) | Ferromagnetic material hardness prediction algorithm based on BP neural network improved algorithm | |
| CN108959175A (en) | A kind of ferrimagnet yield strength successive Regression estimation method based on MBN | |
| de Calle et al. | Dynamic condition monitoring method based on dimensionality reduction techniques for data-limited industrial environments | |
| Yang et al. | Gearbox fault diagnosis based on artificial neural network and genetic algorithms | |
| CN111753751A (en) | Fan fault intelligent diagnosis method for improving firework algorithm | |
| Addeh et al. | Control chart patterns detection using COA based trained MLP neural network and shape features | |
| Miao et al. | Nonlinear model predictive control based on support vector regression | |
| CN114757087A (en) | Tool wear prediction method based on dynamic principal component analysis and LSTM | |
| CN113657244A (en) | Fan gearbox fault diagnosis method and system based on improved EEMD and speech spectrum analysis | |
| Shiri et al. | Using long-term condition monitoring data with non-Gaussian noise for online diagnostics | |
| CN113743010A (en) | Rolling bearing running state evaluation method based on EEMD energy entropy | |
| CN113159225A (en) | Multivariable industrial process fault classification method | |
| Dey et al. | Performance improvement techniques for neural networks in tool condition monitoring | |
| Strasser et al. | An Approach for Adaptive Parameter Setting in Manufacturing Processes. | |
| CN113469408A (en) | Running state trend prediction method and system for phase modulator | |
| Liu et al. | Wind-speed forecasting model based on DBN-Elman combined with improved PSO-HHT | |
| Matei et al. | Classification-based diagnosis using synthetic data from uncertain models | |
| Li et al. | Research on calculating method of hidden layer nodes in BP network |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20211207 |