Injection molding process recognition method based on injection molding machine electrical parameter curve
By collecting electrical parameter curves of injection molding machines, decomposing electrical power using K-Means and CUSUM algorithms, and combining FSST transform and LSTM model, efficient injection molding machine process identification without connection to a controller is achieved. This solves the problems of insufficient data acquisition accuracy and real-time performance in existing technologies, and improves production efficiency and product quality.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SOUTH CHINA UNIV OF TECH
- Filing Date
- 2024-03-06
- Publication Date
- 2026-07-21
AI Technical Summary
Existing technologies suffer from insufficient accuracy and real-time performance in identifying injection molding machine operation processes. Furthermore, they require communication with the injection molding machine controller or the addition of sensors, which poses compatibility and safety risks.
By collecting the electrical parameter curves of the injection molding machine, the electrical power of the motor and heater is decomposed using the K-Means clustering algorithm and the sliding bilateral CUSUM algorithm. Combined with the FSST transform and LSTM network model, the process status of the injection molding machine is identified, achieving identification without the need for connection to the controller.
It improves the accuracy and real-time performance of injection molding machine process identification, simplifies the operation process, reduces data security risks, and improves production efficiency and product quality consistency.
Smart Images
Figure CN118114113B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of injection molding process identification technology, specifically relating to an injection molding process identification method based on injection molding electromechanical parameter curves. Background Technology
[0002] With the transformation and upgrading of enterprises towards informatization and intelligentization, centralized digital management and monitoring of injection molding machines has become a trend. Accurately identifying the processes of an injection molding machine allows for better monitoring and control of the production process, reducing waiting and idle time and improving production efficiency. Secondly, different injection molding processes may require different parameters and conditions; identifying the processes ensures the use of correct process parameters at each stage, improving product quality and consistency. Injection molding machines may experience various malfunctions during operation; identifying the processes makes it easier to locate and diagnose these faults, facilitating timely intervention and reducing production interruptions and maintenance time. Furthermore, identifying the current process allows for targeted energy management, enabling the implementation of corresponding energy-saving measures and reducing production costs.
[0003] Currently, there are two main methods for identifying and acquiring injection molding machine operation data: one is to establish communication with the injection molding machine controller and directly read the operation data; the other is to add external sensors and then identify the process information through data processing. The first method obtains data with high accuracy and real-time performance, but it has problems such as compatibility with the controller's communication interface and the confidentiality of its internal data. The second method requires additional sensors, and the accuracy of the acquired data is closely related to the number and type of sensors, making it complex to implement. Therefore, there is an urgent need to find a method and technology for conveniently identifying and acquiring injection molding machine operation data without intruding on the injection molding machine controller. Summary of the Invention
[0004] The main objective of this invention is to overcome the shortcomings and deficiencies of the prior art and to propose a method for identifying injection molding processes based on injection molding electromechanical parameter curves.
[0005] To achieve the above objectives, the present invention adopts the following technical solution:
[0006] The injection molding process identification method based on injection molding electromechanical parameter curves includes the following steps:
[0007] S1. Set up a data acquisition circuit system to collect the required electrical parameters;
[0008] S2. Based on the power variation characteristics of the injection molding machine motor and heater, the load data of the motor is decomposed according to the electrical power.
[0009] S3. Label and classify the decomposed power time series data according to waveform characteristics;
[0010] S4. Perform FSST transformation on the decomposed power time series data;
[0011] S5. Use LSTM network as process recognition model, train the process recognition model, test the recognition accuracy of the model and use the model for actual injection molding process recognition.
[0012] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0013] 1. The method of this invention integrates multiple collected electrical parameters to identify and decompose the electrical parameter curves of the oil pump motor and heater; based on the decomposed electrical parameters, the process of automatic production of the injection molding machine is identified through the algorithm model, including the process status of mold closing, mold locking, injection, pressure holding, cooling, melting, mold opening, ejection and other processes. The method does not need to be connected to the injection molding machine controller, is simple to operate, and has no data security risks.
[0014] 2. In the part of the invention, the algorithm for the decomposition of the electric power of the motor has been specifically modified. Compared with other technical solutions, such as common change point detection methods based on generalized likelihood ratio test and goodness of fit test, the CUSUM algorithm of the present invention has higher decomposition accuracy and real-time performance.
[0015] 3. In the process identification part, the method of the present invention performs FSST frequency domain transformation on the time series data. Compared with the input of the traditional LSTM model based on one-dimensional time domain data, it has additional frequency domain feature input, which improves the accuracy of the model and thus improves the accuracy of power identification. Attached Figure Description
[0016] Figure 1 This is a flowchart of the method of the present invention;
[0017] Figure 2 This is a schematic diagram of the data acquisition circuit system;
[0018] Figure 3 This is a flowchart of the K-Means clustering algorithm;
[0019] Figure 4 This is a flowchart of the sliding two-sided CUSUM algorithm;
[0020] Figure 5 These are the injection molding machine operating curves collected in the examples;
[0021] Figure 6 This is the power curve of the injection molding machine in the embodiment;
[0022] Figure 7 This is a schematic diagram of the LSTM model;
[0023] Figure 8 This is a diagram illustrating the accuracy and loss value of the LSTM model obtained after training.
[0024] Figure 9 This is a schematic diagram illustrating the accuracy of the LSTM model in identifying process steps from frequency domain input data. Detailed Implementation
[0025] The present invention will be further described in detail below with reference to the embodiments and accompanying drawings, but the embodiments of the present invention are not limited thereto.
[0026] Example
[0027] like Figure 1 As shown, the present invention provides a method for identifying injection molding processes based on injection molding electromechanical parameter curves, comprising the following steps:
[0028] S1. Set up a data acquisition circuit system to collect the required electrical parameters;
[0029] like Figure 2 As shown, specifically:
[0030] A data acquisition circuit system is constructed, which includes an electrical parameter acquisition module and current and voltage transformers.
[0031] Connect the current and voltage transformers to the injection molding machine's input side to collect the total load power of the injection molding machine and obtain the voltage and current of the injection molding machine's bus in real time;
[0032] The data acquisition circuit system software was developed using the C# language, and the Modbus TCP communication protocol was used to read various electrical parameters (including current, voltage, power and power factor) from the electrical parameter acquisition module.
[0033] S2. Based on the power variation characteristics of the injection molding machine motor and heater, and combined with the electrical power decomposition, the motor load data is derived, including:
[0034] S21. The electrical power consumed by the heater is obtained using a one-dimensional K-Means clustering algorithm, specifically:
[0035] The one-dimensional K-Means clustering algorithm is used to first cluster the load data of the heating section of the injection molding machine to obtain the power values during heating and when the injection molding machine is in standby mode, thereby obtaining the power consumed by the heater;
[0036] First, the data collected during the preheating stage of the injection molding machine at the start of operation are clustered using the K-means algorithm to obtain two types of results: the working power value of the heater and the standby power value of the injection molding machine. The difference between the two types is calculated and the absolute value is taken to obtain the preliminary value of the power difference required for determination.
[0037] When performing clustering, the input is one-dimensional power data, and the number of clusters is fixed at 2. Using a simplified form, the K-Means clustering algorithm takes the electric power data of the motor and heater measured by the bus as input and outputs the difference in heater power.
[0038] like Figure 3 As shown, the specific process of the K-Means clustering algorithm is as follows:
[0039] After inputting multiple data points, each sample point is first assigned to the initial center point closest to it. Then, the center point of each cluster is calculated and replaced with the original center point. This process is repeated iteratively until the center point remains unchanged. The algorithm then terminates and outputs the results.
[0040] S22. The sliding bilateral CUSUM algorithm performs event detection based on the obtained heater power difference, specifically as follows:
[0041] Event detection is performed using a sliding bilateral CUSUM algorithm. A scanning approach is adopted, merging the mean window and the transient window. The window length is d. The mean is calculated as the mean of the k-1 points preceding the k-th point in the window, using the statistical function calculation method in the non-parametric CUSUM algorithm. The expression is:
[0042]
[0043]
[0044] Where k is the k-th point in the window, d k The value at the current point is σ, and the noise threshold is σ. Changes in the absolute value within σ are negligible.
[0045] The cumulative sum within the window is calculated using the above formula. When the cumulative sum exceeds the set cumulative sum threshold, it is determined that an event has occurred.
[0046] like Figure 4 As shown, the specific process of the sliding CUSUM algorithm is as follows:
[0047] First, determine the parameters, including the noise threshold, cumulative sum threshold, and window length.
[0048] Event detection is performed. When the cumulative sum does not reach the threshold, k is judged. When k is greater than the window length, the window moves one unit to the right. When k is less than the window length, k is incremented by 1 and the accumulation continues. When an event is determined to have occurred, the window moves k units to the right according to the value of k, and the value of k is updated to 0.
[0049] S23. Separate the power values of the heater and the motor. Specifically, select the power value of the heater and the power factor obtained by K-means clustering from the electrical parameters (including current, voltage, power and power factor) collected in step S1, and use the sliding bilateral CUSUM algorithm to separate the power values of the heater and the motor.
[0050] like Figure 5 The figure shows the working curve of the injection molding machine collected in the embodiment.
[0051] S3. The decomposed power time-series data are labeled and classified according to waveform characteristics, specifically as follows:
[0052] Based on prior knowledge, the decomposed one-dimensional time-series data of electric power is labeled and classified according to waveform characteristics. According to the working conditions of the injection molding machine, it is divided into eight categories: mold closing, mold locking, injection, pressure holding, melting, mold opening, ejector pin ejection, and null.
[0053] By labeling the starting point range corresponding to each working condition in the decomposed one-dimensional time-series data samples of electric power, regional labels are formed, enabling accurate positioning of process waveforms based on the regional range.
[0054] like Figure 6 The figure shows the power curve of the injection molding machine in the embodiment.
[0055] S4. Perform FSST transformation on the decomposed power time series data, specifically as follows:
[0056] The FSST Fourier synchronous compression transform is used to perform time-frequency analysis on the one-dimensional time series data of electric power, converting the one-dimensional time series data of electric power into time-frequency data;
[0057] In the FSST Fourier Synchronous Compression Transform, the signal is first subjected to a Fourier transform to convert the time-domain signal into a frequency-domain representation, and then a synchronization compression technique is applied in the frequency domain.
[0058] For time-frequency analysis, the sum of the analytic signal can be expressed as:
[0059]
[0060] Where, φ k (t) has the time derivative dφ corresponding to the instantaneous frequency. k (t) / dt;
[0061] The Fourier synchronous compression transform is based on the short-time Fourier transform implemented in a spectral function, which uses a spectral window g and computes the short-time Fourier transform of the deterministic function f:
[0062]
[0063] The transform value is compressed, concentrating it around the instantaneous frequency curve on the time-frequency plane. The resulting synchronous compressed transform form is:
[0064]
[0065] S5. Construct a process recognition model based on LSTM network, train the process recognition model, test the recognition accuracy of the model, and use the model for actual injection molding process recognition.
[0066] Among them, such as Figure 7 As shown, the model structure of the LSTM network includes:
[0067] Cell state, maintaining a long-term internal state for transmitting and storing information. The cell state is updated at each time step;
[0068] The hidden states, along with the cell states, constitute the output of the LSTM and are used to convey information about the network. The hidden states are also updated at each time step.
[0069] The forget gate determines which information to discard from the cell state. It receives the hidden state from the previous time step and the current input, and outputs a value between 0 and 1.
[0070] The input gate determines which information to store from the current input. Similar to the forget gate, it takes the hidden state from the previous time step and the current input, and outputs a value between 0 and 1.
[0071] The output gate determines what information to output to the hidden state at the next time step. It receives the hidden state at the previous time step and the current input, and outputs a value between 0 and 1.
[0072] The LSTM network first takes a sequence of data as input and provides an initial hidden state and cell state, which are usually initialized as zero vectors. Then, at each time step, the cell state and hidden state are updated through the calculation of the forget gate, input gate, and output gate. Subsequently, the corresponding output is generated at each time step of the sequence. Finally, at the end of the sequence, the hidden state or cell state of the last time step can be selected as the final output.
[0073] Step S5 is as follows:
[0074] By modifying the feature dimension of the input layer in the LSTM network architecture, the network accepts the spectrum of each sample instead of a single value. The frequency domain data after FSST transformation and the corresponding working condition label are used as feature inputs into the LSTM network for training.
[0075] During training, the LSTM network learns spectral features from the process sequence of the injection molding cycle of the injection molding machine and uses them to perform specific classification tasks to obtain the overall accuracy and loss value of the LSTM network. If the accuracy and loss value do not meet the preset requirements, the number of training iterations and the learning rate of the LSTM network are adjusted, and iterative learning continues until the accuracy and loss value meet the preset requirements.
[0076] Finally, the trained LSTM network was used for actual injection molding process recognition.
[0077] like Figure 8 The image shows a diagram illustrating the accuracy and loss values of the trained LSTM model; as shown... Figure 9 The figure shows the accuracy of the LSTM model in identifying the process of the frequency domain input data.
[0078] It should also be noted that, in this specification, terms such as "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0079] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for identifying injection molding processes based on injection molding electromechanical parameter curves, characterized in that, Includes the following steps: S1. Set up a data acquisition circuit system to collect the required electrical parameters; S2. Based on the power variation characteristics of the injection molding machine motor and heater, and combined with the electrical power decomposition, the motor load data is derived, specifically including: S21. Use the one-dimensional K-Means clustering algorithm to obtain the electrical power consumed by the heater; S22. The sliding bilateral CUSUM algorithm performs event detection based on the obtained difference in heater power. S23, the electrical power values of the separate heater and motor; S3. Label and classify the decomposed power time series data according to waveform characteristics; S4. Perform FSST transformation on the decomposed power time series data; S5. Use an LSTM network as the process identification model, train the process identification model, test the model's recognition accuracy, and apply the model to actual injection molding process identification. Specifically, by modifying the feature dimension of the input layer in the LSTM network architecture, the network accepts the spectrum of each sample instead of a single value. The frequency domain data after FSST conversion and the corresponding working condition label are used as features input into the LSTM network for training.
2. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 1, characterized in that, Step S1 is as follows: A data acquisition circuit system is constructed, which includes an electrical parameter acquisition module and current and voltage transformers. Connect the current and voltage transformers to the injection molding machine's input side to collect the total load power of the injection molding machine and obtain the voltage and current of the injection molding machine's bus in real time; The data acquisition circuit system software was developed using the C# language, and the Modbus TCP communication protocol was used to read various electrical parameter data from the electrical parameter acquisition module.
3. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 1, characterized in that, Step S21 is as follows: The one-dimensional K-Means clustering algorithm is used to first cluster the load data of the heating section of the injection molding machine to obtain the power values during heating and when the injection molding machine is in standby mode, thereby obtaining the power consumed by the heater. First, the data collected during the preheating stage of the injection molding machine at the start of operation are clustered using the K-means algorithm to obtain two types of results: the working power value of the heater and the standby power value of the injection molding machine. The difference between the two types is calculated and the absolute value is taken to obtain the preliminary value of the power difference required for determination. When performing clustering, the input is one-dimensional power data, and the number of clusters is fixed at 2. Using a simplified form, the K-Means clustering algorithm takes the electric power data of the motor and heater measured by the bus as input and outputs the difference in heater power. The specific process of the K-Means clustering algorithm is as follows: After inputting multiple data points, each sample point is first assigned to the initial center point closest to it. Then, the center point of each cluster is calculated and replaced with the original center point. This process is repeated iteratively until the center point remains unchanged. The algorithm then terminates and outputs the results.
4. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 1, characterized in that, Step S22 is as follows: Event detection is performed using a sliding bilateral CUSUM algorithm. A scanning approach is adopted, merging the mean window and the transient window. The window length is d. The mean is calculated as the mean of the k-1 points preceding the k-th point in the window, using the statistical function calculation method in the non-parametric CUSUM algorithm. The expression is: ； ； Where k is the k-th point in the window, The value at the current point. The noise threshold is defined as follows: The changes within are negligible; The cumulative sum within the window is calculated using the above formula. When the cumulative sum exceeds the set cumulative sum threshold, it is determined that an event has occurred. The specific process of the sliding two-sided CUSUM algorithm is as follows: First, determine the parameters, including the noise threshold, cumulative sum threshold, and window length. Event detection is performed. When the cumulative sum does not reach the threshold, k is judged. When k is greater than the window length, the window moves one unit to the right. When k is less than the window length, k is incremented by 1 and the accumulation continues. When an event is determined to have occurred, the window moves k units to the right according to the value of k, and the value of k is updated to 0.
5. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 3, characterized in that, Step S23 is as follows: In the electrical parameters collected in step S1, the power value of the heater and the power factor characteristics obtained by K-means clustering are selected. The electric power is separated by the sliding bilateral CUSUM algorithm to separate the electric power values of the heater and the motor.
6. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 1, characterized in that, Step S3 is as follows: Based on prior knowledge, the decomposed one-dimensional time-series data of electric power is labeled and classified according to waveform characteristics. According to the working conditions of the injection molding machine, it is divided into eight categories: mold closing, mold locking, injection, pressure holding, melting, mold opening, ejector pin ejection, and null. By labeling the starting point range corresponding to each working condition in the decomposed one-dimensional time-series data samples of electric power, regional labels are formed, enabling accurate positioning of process waveforms based on the regional range.
7. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 1, characterized in that, Step S4 is as follows: The FSST Fourier synchronous compression transform is used to perform time-frequency analysis on the one-dimensional time series data of electric power, converting the one-dimensional time series data of electric power into time-frequency data; In the FSST Fourier Synchronous Compression Transform, the signal is first subjected to a Fourier transform to convert the time-domain signal into a frequency-domain representation, and then a synchronization compression technique is applied in the frequency domain. For time-frequency analysis, the sum of the analytic signal is expressed as: in, It has a time derivative corresponding to the instantaneous frequency. ; The Fourier synchronous compression transform is based on the short-time Fourier transform implemented in a spectral function, which uses a spectral window g and computes the short-time Fourier transform of the deterministic function f: The transform value is compressed, concentrating it around the instantaneous frequency curve on the time-frequency plane. The resulting synchronous compressed transform form is:
8. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 1, characterized in that, In step S5, the model structure of the LSTM network includes: Cell state, which maintains a long-term internal state for transmitting and storing information, is updated at each time step; The hidden state, together with the cell state, constitutes the output of the LSTM and is used to convey information about the network. The hidden state is also updated at each time step. The forget gate determines which information to discard from the cell state. It receives the hidden state from the previous time step and the current input, and outputs a value between 0 and 1. The input gate determines which information to store from the current input, similar to the forget gate. It takes the hidden state from the previous time step and the current input, and outputs a value between 0 and 1. The output gate determines what information to output to the hidden state at the next time step. It receives the hidden state at the previous time step and the current input, and outputs a value between 0 and 1. The LSTM network first takes a sequence of data as input and provides an initial hidden state and cell state, which are usually initialized as zero vectors. Then, at each time step, the cell state and hidden state are updated through the calculation of the forget gate, input gate, and output gate. Subsequently, the corresponding output is generated at each time step of the sequence. Finally, at the end of the sequence, the hidden state or cell state of the last time step can be selected as the final output.
9. The injection molding process identification method based on injection molding electromechanical parameter curves according to claim 8, characterized in that, Step S5 is as follows: During training, the LSTM network learns spectral features from the process sequence of the injection molding cycle of the injection molding machine and uses them to perform specific classification tasks to obtain the overall accuracy and loss value of the LSTM network. If the accuracy and loss value do not meet the preset requirements, the number of training iterations and the learning rate of the LSTM network are adjusted, and iterative learning continues until the accuracy and loss value meet the preset requirements. Finally, the trained LSTM network was used for actual injection molding process recognition.