A monitoring method and device for abnormal injection molding processes of an injection molding machine

Abstract

The present invention belongs to the technical field related to the monitoring of injection molding machines, and discloses a method and device for monitoring abnormal injection molding processes, including two stages: offline training and online monitoring. In the offline training stage, characteristic data and pressure-position data under various different process conditions are collected; the characteristic data are combined into a characteristic data set, and the pressure-position data are used to construct a feature extraction model to realize feature extraction of the pressure-position data; a process recognition model is constructed using the dimensionality-reduced vectors output by the feature extraction model. In the online monitoring stage, the dimensionality-reduced vectors, error vectors, and collected characteristic data output by the feature extraction model are combined into feature vectors; the dimensionality-reduced vectors are input into the process recognition model to obtain the process category to which they belong, and the historical feature vectors corresponding to the corresponding process are obtained according to the process category. The historical feature vectors are reused and the anomaly factor is calculated based on the anomaly detection method, and then anomaly monitoring is carried out according to the anomaly factor. The present invention improves the accuracy rate.

Description

Technical Field

[0001] The present invention belongs to the technical field related to the monitoring of injection molding machines, and more specifically, relates to a method and device for monitoring abnormalities in the injection molding process of an injection molding machine. Background Art

[0002] During the production process of injection molded products, due to interference from the external environment, the product quality will change. Discovering abnormalities in production and making timely adjustments can improve the yield rate. As the injection molding industry gradually enters the field of precision devices, the requirements for the molding process are becoming more and more stringent. The injection molding production process is becoming more and more sensitive and more vulnerable to external interference. The main requirement of the enterprise mass production department is to monitor the production process in real time and timely remind the process personnel to adjust the process and troubleshoot the reasons.

[0003] The injection molding process is a black box process. At present, process personnel basically observe through simple characteristic data (peak pressure, injection time, etc.) provided by the injection molding machine. Monitoring is carried out by setting upper and lower thresholds for individual characteristic data, and such simple characteristic data is difficult to comprehensively and effectively characterize the production process and completely relies on the experience of process personnel. There are many feature extraction methods for pressure-position curves in the academic community, but most of the feature extraction methods only consider the feature extraction process and ignore the reconstruction process. Summary of the Invention

[0004] In view of the above defects or improvement requirements of the prior art, the present invention provides a method and device for monitoring abnormalities in the injection molding process of an injection molding machine. The method includes two stages: offline training and online monitoring. In the offline training stage, characteristic data and pressure-position data under various different process conditions are collected; the characteristic data is combined into a characteristic data set, and the pressure-position data is used to construct a feature extraction model to realize feature extraction of the pressure-position data; a process recognition model is constructed using the reduced-dimensional vectors output by the feature extraction model. In the online monitoring stage, the reduced-dimensional vectors, error vectors, and collected characteristic data output by the feature extraction model are combined into a feature vector; the reduced-dimensional vectors are input into the process recognition model to obtain the process category to which they belong, and the historical feature vectors corresponding to the corresponding process are obtained according to the process category. The historical feature vectors are input into the anomaly detection method to calculate the anomaly factor, and it is judged whether it is abnormal according to whether the anomaly factor is greater than the set threshold. If it is abnormal, the injection molding machine is notified to alarm, thereby effectively monitoring the injection molding process.

[0005] To achieve the above object, according to one aspect of the present invention, a method for monitoring abnormalities in the injection molding process of an injection molding machine is provided. The method mainly includes the following steps:

[0006] (1) Collect pressure-position data and corresponding characteristic data under multiple process conditions and for multiple die passes, and preprocess the multiple groups of pressure-position data to obtain a process data set, and the corresponding combination of characteristic data is the characteristic data set;

[0007] (2) Use the process data set to construct a feature extraction model, and obtain a dimensionality-reduced vector and an error vector based on the process data set and the constructed feature extraction model;

[0008] (3) Combine the dimensionality-reduced vectors, error vectors, and characteristic data of all die passes in the process data set into a group of feature vectors, and then obtain a feature vector set. The dimensionality-reduced vectors and error vectors are combined into process vectors, and then a process vector set is obtained. And construct a process identification model based on the dimensionality-reduced vectors in the process vector set;

[0009] (4) In the online monitoring stage, preprocess the collected pressure-position data and input it into the feature extraction model to obtain a dimensionality-reduced vector, and further obtain an error vector. The dimensionality-reduced vector, error vector, and characteristic data form a new feature vector. Input the obtained dimensionality-reduced vector into the process identification model to obtain the current process to which it belongs, obtain the historical feature vector corresponding to the current process in the feature vector set, insert the new feature vector into the historical feature vector, and calculate the anomaly factor of the current die pass based on the obtained historical feature vector using an anomaly detection method, and then perform anomaly judgment based on this anomaly factor.

[0010] Further, for a process, continuously produce n die passes under stable process conditions, record the pressure-position data of n die passes, and preprocess the pressure-position data and convert it into a matrix N:

[0011] N = {N1, N2,..., N n-1 , N n}

[0012] N i = {P, Q}

[0013] P = {p1, p2,..., p k-1 , p k}

[0014] Q = {q1, q2,..., q k-1 , q k}

[0015] where n is the number of die passes, p i is the injection pressure value, q i is the screw position value, k is the number of sampling points, and the characteristic data of n die passes are collected at the same time.

[0016] Further, data under m different process conditions are collected in total to form a process data set M: M = {M1, M2,..., M m-1 , M m}, and the collected feature data forms a feature data set.

[0017] Further, the feature extraction model includes an encoder and a decoder; the process data set is used as the input of the feature extraction model. When the reconstruction error of the feature extraction model is lower than the set threshold, the training ends, and the constructed process parameter feature extraction model is saved.

[0018] Further, the pressure-position data is input into the encoder in the process parameter feature extraction model to obtain a dimensionality-reduced vector, and the dimensionality-reduced vector is input into the decoder in the process parameter feature extraction model to obtain reconstructed data. The error vector is calculated from the reconstructed data and the original data; a process identification model is constructed based on the dimensionality-reduced vector set in the feature vector set.

[0019] Further, the self-learning model uses a variational autoencoder. The neural networks of the encoder and the decoder use two-layer long short-term memory artificial neural networks, and the last layer of the decoder uses a fully connected network.

[0020] Further, the parameters of the neural network are updated using the RMSprop algorithm.

[0021] Further, the dimensionality-reduced vector, the error vector, and the feature data are combined into a group of feature vectors; this is repeated m×n times to obtain a dimensionality-reduced vector set C composed of dimensionality-reduced vectors with a dimension of m×n×dc, where dc is the dimension of the dimensionality-reduced vector, and a feature vector set with a dimension of m×n×da is obtained, where da is the dimension of the feature vector.

[0022] The present invention provides a monitoring system for abnormal injection molding processes of an injection molding machine. The monitoring system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it executes the monitoring method for abnormal injection molding processes of the injection molding machine as described above.

[0023] The present invention provides a computer-readable storage medium. The computer-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by a processor, the machine-executable instructions cause the processor to implement the monitoring method for abnormal injection molding processes of the injection molding machine as described above.

[0024] Generally speaking, compared with the prior art through the above technical solutions conceived by the present invention, the monitoring method and device for abnormal injection molding processes provided by the present invention mainly have the following beneficial effects:

[0025] 1. Most feature extraction methods only consider the feature extraction process while ignoring the reconstruction process. This method calculates the error vector based on the reconstructed data and the original data and uses it for further calculation of the anomaly factor.

[0026] 2. Using only the pressure-position data as the input of the monitoring method is still insufficient to describe the forming process. Therefore, this method introduces the simple feature data provided by the injection molding machine and combines the feature data set to make the description more comprehensive.

[0027] 3. In the process of calculating the anomaly factor, considering the inconsistent data distribution under different process conditions, a process recognition model is constructed using the dimensionality-reduced vector, and the historical feature vectors corresponding to the process are reused. The distribution of the current historical feature vectors is integrated in the link of calculating the anomaly factor to make the calculated anomaly factor more accurate.

[0028] 4. The present invention performs feature extraction through a variational autoencoder model based on a long short-term memory artificial neural network. The encoder of the variational autoencoder extracts features from the data of the forming process to obtain a dimensionality-reduced vector, and the decoder of the variational autoencoder performs reconstruction to calculate the reconstructed error vector, and finally combines them into a feature vector. This method takes into account the autocorrelation and cross-correlation of the process data and can effectively perform feature extraction.

[0029] 5. The present invention calculates the scores of the feature vectors describing each mold cycle through a density-based local outlier detection algorithm. Considering the continuity of injection molding, it evaluates each mold cycle more accurately and improves the sensitivity of detecting anomalies.

[0030] 6. The present invention makes the feature extraction have strong generalization ability by introducing an autoencoder model. Under the working conditions of new process parameters, there is no need to retrain, and the whole process is efficient and accurate without introducing artificial experience for feature extraction. Description of the Drawings

[0031] Figure 1 is a schematic flowchart of a method for monitoring anomalies in the injection molding process of an injection molding machine provided by the present invention;

[0032] Figure 2 is an injection pressure curve graph under different process conditions;

[0033] Figure 3 is a graph of the change of loss during the training process of the feature extraction model;

[0034] Figure 4 is a comparison graph of the similarity ratio of the feature extraction model compared with the PCA-based method;

[0035] Figure 5 is a schematic diagram of calculating the anomaly factor score using this example in production applications. Specific implementation mode

[0036] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. In addition, the technical features involved in the various implementation modes of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0037] Please refer to Figure 1 , the present invention provides a monitoring method for abnormal injection molding processes of an injection molding machine, and the monitoring method mainly includes the following steps:

[0038] Step 1, collect pressure-position data and corresponding characteristic data under various process conditions and for multiple molding cycles, and preprocess the multiple sets of pressure-position data to obtain a process data set, and combine the corresponding characteristic data to obtain a characteristic data set.

[0039] In this implementation mode, for one process, continuously produce n molding cycles under stable working conditions, record the pressure-position data and characteristic data of n molding cycles, and the collected characteristic data forms a characteristic data set. After preprocessing the pressure-position data, it is converted into a matrix N:

[0040] N = {N1, N2,..., N n-1 , N n}

[0041] N i = {P, Q}

[0042] P = {p1, p2,..., p k-1 , p k}

[0043] Q = {q1, q2,..., q k-1 , q k}

[0044] Where n is the number of molding cycles, p i is the injection pressure value, q i is the screw position value, and k is the number of sampling points

[0045] Repeat the above steps under m different process conditions to obtain the corresponding matrix M i , M i constitutes a process data set M: M = {M1, M2,..., M m-1 , M m}.

[0046] The preprocessing is specifically as follows: The pressure-position data is divided according to the injection molding stage, and the pressure injection section and the position holding section are retained. After K samplings respectively, the pressure injection section PL and the position holding section QL are obtained. PL and QL are normalized according to the maximum and minimum values to obtain P and Q:

[0047]

[0048] where PL_max is the maximum value of the PL vector, PL_min is the minimum value of the PL vector, the QL vector is processed in the same way as the P vector to obtain the vector Q, and the process data N is obtained after processing according to the above process. i .

[0049] Step 2: Use the process data set to train the feature extraction model, and obtain the dimensionality reduction vector and the error vector in turn based on the process data set and the trained feature extraction model.

[0050] The feature extraction model includes an encoder and a decoder. The process data set is used as the input of the feature extraction model, and the feature extraction model is saved after the reconstruction error is lower than the set threshold.

[0051] Input M ij ={P, Q} in the process data set into the encoder in the feature extraction model to obtain the dimensionality reduction vector, input the dimensionality reduction vector into the decoder in the feature extraction model to obtain the reconstructed data, and calculate the error vector from the reconstructed data and the original data, where 1 ≤ i ≤ m and 1 ≤ j ≤ n.

[0052] In this embodiment, the self-learning model uses a variational autoencoder. The neural networks of the encoder and the decoder use two-layer long short-term memory artificial neural networks, and the last layer of the decoder uses a fully connected network; the parameters of the neural network are updated using the RMSprop algorithm.

[0053] Step 3: Combine the dimensionality reduction vectors, error vectors, and feature data of all shots in the process data set into a set of feature vectors, and then obtain a feature vector set, and construct a process recognition model based on the dimensionality reduction vectors in the process vector set;

[0054] The dimensionality reduction vector, the error vector, and the feature data are combined into a set of feature vectors; repeat m × n times to obtain a dimensionality reduction vector set C composed of dimensionality reduction vectors with a dimension of m × n × dc, where dc is the dimension of the dimensionality reduction vector, and obtain a feature vector set with a dimension of m × n × da, where da is the dimension of the feature vector.

[0055] In the dimensionality reduction vector set C ij, where \(1\leq i\leq m\), \(1\leq j\leq n\), the label is set to \(i\) to obtain the label, and the length of the label is \(m\times n\). The dimensionality-reduced vector set and the label are used as the input of the process recognition model to train a process recognition model capable of distinguishing different working conditions, and the process recognition model is saved after training is completed.

[0056] In this embodiment, the process recognition model adopts the K-means unsupervised learning method.

[0057] Step 4, in the online monitoring stage, the collected pressure-position data is preprocessed and then input into the feature extraction model to obtain the dimensionality-reduced vector, and then the error vector is obtained. The dimensionality-reduced vector, the error vector and the feature data form a new feature vector. The obtained dimensionality-reduced vector is input into the process recognition model to obtain the current process to which it belongs, and the historical feature vector corresponding to the current process in the feature vector set is obtained. The new feature vector is inserted into the historical feature vector, and the anomaly factor of the current die is calculated based on the obtained historical feature vector, and then the anomaly judgment is performed based on the anomaly factor.

[0058] Among them, an anomaly detection algorithm is used to calculate the new feature vector to obtain the anomaly factor, and it is judged whether an alarm needs to be sent to the injection molding machine by whether the anomaly factor is greater than the set threshold.

[0059] In this embodiment, the density-based local outlier factor (LOF) is used to calculate the anomaly factor.

[0060] As Figure 2 shown, it is a single-group injection section pressure curve obtained under six different process conditions after sampling and aligning \(K\) times, which has not been normalized yet, where \(K\) in this example is 250.

[0061] As Figure 3 shown, during the training process, the neural network optimization and update method adopts the RMSprop algorithm, the learning rate is initially set to 0.001, and the cosine annealing is used to control the learning rate and the loss change during the training process.

[0062] As Figure 4 shown, the performance of the feature extraction model using the autoencoder is significantly stronger than the traditional PCA method. The figure shows the similarity ratio between the reconstructed data and the original data of the injection section pressure data. The left side is the method based on VAE, and the right side is the method based on PCA.

[0063] In this example, the similarity ratio of the Euclidean distance is used to evaluate the similarity degree between the original data and the reconstructed data. When using the Euclidean distance to represent the similarity, it is converted through the following formula:

[0064] sim(X,Y)=1 / (1+d(X,Y))

[0065] Where d(X,Y) is the Euclidean distance. Compared with the PCA method, the method of this example can better mine the autocorrelation and dynamic correlation of process data and has stronger non-linear expression ability.

[0066] As Figure 5 shown, for the abnormal factor score of on-site products, the threshold can be set to 2. When it is greater than 2, an alarm signal is sent. A sliding window is used to record the number of alarms received in a recent period of time. When it is greater than the set number, the process personnel are reminded.

[0067] This embodiment takes into account the autocorrelation and dynamic correlation of process data, and also considers the inconsistent data distributions under different process conditions when calculating the abnormal factor, can quickly and sensitively sense the abnormalities in the injection molding process and notify the alarm, and is applicable to industrial applications.

[0068] The present invention also provides a monitoring system for abnormalities in the injection molding process of an injection molding machine. The monitoring system includes a memory and a processor. The memory stores a computer program. When the processor executes the computer program, it executes the monitoring method for abnormalities in the injection molding process of the injection molding machine as described above.

[0069] The present invention also provides a computer-readable storage medium. The computer-readable storage medium stores machine-executable instructions. When the machine-executable instructions are called and executed by the processor, the machine-executable instructions cause the processor to implement the monitoring method for abnormalities in the injection molding process of the injection molding machine as described above.

[0070] It is easy for those skilled in the art to understand that the above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A monitoring method for abnormal injection molding processes of an injection molding machine, characterized in that, The method comprises the following steps: (1) Collect pressure-position data and corresponding characteristic data under multiple process conditions and for multiple shots, and preprocess the multiple groups of pressure-position data to obtain a process data set, and the corresponding combination of characteristic data is a characteristic data set; (2) Use the process data set to construct a feature extraction model, and obtain a dimensionality-reduced vector and an error vector based on the process data set and the constructed feature extraction model; (3) Combine the dimensionality-reduced vectors, error vectors, and characteristic data of all shots in the process data set into a group of feature vectors, and then obtain a feature vector set. The dimensionality-reduced vectors and error vectors are combined into a process vector, and then a process vector set is obtained. And construct a process identification model based on the dimensionality-reduced vectors in the process vector set; (4) In the online monitoring stage, preprocess the collected pressure-position data and input it into the feature extraction model to obtain a dimensionality-reduced vector, and then obtain an error vector. The dimensionality-reduced vector, error vector, and characteristic data form a new feature vector. Input the obtained dimensionality-reduced vector into the process identification model to obtain the current process to which it belongs, obtain the historical feature vector corresponding to the current process in the feature vector set, insert the new feature vector into the historical feature vector, and calculate the anomaly factor of the current shot using an anomaly detection method based on the obtained historical feature vector, and then perform anomaly judgment based on the anomaly factor.

2. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 1, characterized in that: For a certain process, continuously produce n shots under stable process conditions, record the pressure-position data of n shots, and preprocess the pressure-position data and convert it into a matrix N: N = {N1, N2,..., N n-1 , N n} N i = {P, Q} P = {p1, p2,..., p k-1 , p k} Q = {q1, q2,..., q k-1 , q k} where n is the number of molding cycles, p i is the injection pressure value, q i is the screw position value, k is the number of sampling points, and the characteristic data of n molding cycles are collected simultaneously.

3. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 2, characterized in that: A total of m types of data under different process conditions are collected to form a process data set M: M = {M1, M2,..., M m-1 , M m}, and the simultaneously collected characteristic data forms a characteristic data set.

4. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 1, characterized in that: The feature extraction model includes two parts: an encoder and a decoder; use the process data set as the input of the feature extraction model, and end the training when the reconstruction error of the feature extraction model is lower than the set threshold, and save the constructed process parameter feature extraction model.

5. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 4, characterized in that: Input the pressure-position data into the encoder in the process parameter feature extraction model to obtain a dimensionality-reduced vector, input the dimensionality-reduced vector into the decoder in the process parameter feature extraction model to obtain reconstructed data, and calculate the error vector from the reconstructed data and the original data; Construct a process identification model based on the dimensionality-reduced vector set in the feature vector set.

6. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 4, characterized in that: The self-learning model uses a variational autoencoder, and the neural networks of the encoder and decoder use two-layer long short-term memory artificial neural networks, and the last layer of the decoder uses a fully connected network.

7. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 6, characterized in that: The parameters of the neural network are updated using the RMSprop algorithm.

8. The monitoring method for abnormal injection molding process of an injection molding machine according to claim 5, characterized in that: The dimensionality-reduced vector, error vector, and characteristic data are combined into a group of feature vectors; repeat m×n times to obtain a dimensionality-reduced vector set C composed of dimensionality-reduced vectors with a dimension of m×n×dc, where dc is the dimension of the dimensionality-reduced vector, and obtain a feature vector set with a dimension of m×n×da, where da is the dimension of the feature vector.

9. A monitoring system for abnormal injection molding processes of an injection molding machine, characterized in that: The monitoring system includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it executes the monitoring method for the abnormality of the injection molding process of the injection molding machine according to any one of claims 1-8.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores machine-executable instructions, and when the machine-executable instructions are called and executed by a processor, the machine-executable instructions cause the processor to implement the monitoring method for abnormal injection molding processes of the injection molding machine according to any one of claims 1-8.

Images