Injection molding process data extraction method based on oil pump motor electric parameter signals

Abstract

The invention discloses an oil pump motor electrical parameter signal-based injection molding process data extraction method, which comprises the following steps of S1, collecting power curve data of an injection molding machine, and performing data preprocessing; s2, constructing a BiLSTM-Attention network, and carrying out iterative training on the BiLSTM-Attention network; s3, constructing a finite-state machine based on an injection molding process; s4, fusing the trained BiLSTM-Attention network and a finite-state machine based on an injection molding process, identifying the injection molding process, and generating an identification result curve with a process label; and S5, extracting process data of each process according to the identification result curve with the process label. According to the method, a deep learning network BiLSTM-Attention is established, an injection molding process state machine based on injection molding professional knowledge is introduced, the deep learning network and the injection molding process state machine are fused to carry out injection molding process identification, and finally accurate extraction of injection molding process data is realized.

Description

Technical Field

[0001] The present invention belongs to the technical field of injection molding machines, and particularly relates to a method for extracting injection molding process data based on the electrical parameter signals of an oil pump motor. Background Art

[0002] With the transformation and upgrading of digitalization and intelligentization, predictive maintenance and real-time data analysis methods based on artificial intelligence play a more important role in the field of injection molding machines. Using the electrical parameter signals of the oil pump motor for digital research is more convenient than the traditional internal sensor signals of injection molding machines. It can identify the process and diagnose some common faults while monitoring energy consumption. Due to the step-by-step and periodic working characteristics of injection molding machines, the active power of the injection molding machine oil pump motor directly reflects the working state and load of the hydraulic system, which are closely related to the specific process being carried out by the injection molding machine. Therefore, the electrical parameter signals of the injection molding machine oil pump motor can be used to identify the process of the injection molding machine.

[0003] In the prior art, Patent CN118114113A discloses a method for identifying injection molding processes based on the electrical parameter curve of an injection molding machine. By extracting the periodic process signals of the injection molding machine, processing the data, and then passing it through an LSTM (Long Short-Term Memory neural network), the state at each time point is obtained, and then the process situation is obtained through the state at each time point. However, this method simply passes the electrical parameter signals of the oil pump motor through an LSTM model, resulting in a low accuracy rate for process identification and there are also cases of misidentification, making it very difficult to divide the process based on this method. Summary of the Invention

[0004] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and propose a method for extracting injection molding process data based on the electrical parameter signals of an oil pump motor.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A method for extracting injection molding process data based on the electrical parameter signals of an oil pump motor, comprising the following steps:

[0007] S1. Collect the power curve data of the injection molding machine and perform data preprocessing;

[0008] S2. Build a BiLSTM-Attenion network and perform iterative training on it;

[0009] S3. Construct a finite state machine based on the injection molding process;

[0010] S4. Integrate the trained BiLSTM-Attenion network and the finite state machine based on the injection molding process to identify the injection molding process and generate an identification result curve with process labels;

[0011] S5. Extract the process data of each process according to the recognition result curve with process labels.

[0012] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0013] 1. The present invention constructs an excellent deep learning model BiLSTM-Attention for identifying the injection molding process based on the electrical parameters of the oil pump motor. Compared with the existing models for process identification, this model has a higher accuracy. The present invention also introduces an injection molding process state machine based on injection molding professional knowledge to determine the authenticity of the deep learning-identified process results and correct misjudged processes, thereby completing the accurate identification of process data. By integrating the deep learning network and the injection molding process state machine for injection molding process identification, the accurate extraction of injection molding process data is ultimately achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flowchart of the method of the present invention;

[0015] Figure 2 is a schematic diagram of the data preprocessing step in the embodiment;

[0016] Figure 3 is a schematic diagram of the BiLSTM-Attention network structure constructed in the embodiment;

[0017] Figure 4 is a diagram of the injection molding process recognition result of the trained BiLSTM-Attention network in the embodiment;

[0018] Figure 5 is a schematic diagram of the finite state machine based on the injection molding process;

[0019] Figure 6 is a schematic diagram of the injection molding process recognition result of the single BiLSTM-Attention network in the embodiment;

[0020] Figure 7 is a schematic diagram of the injection molding process recognition result of the fusion model in the embodiment. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] The present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings, but the embodiments of the present invention are not limited thereto.

[0022] Embodiment

[0023] As Figure 1 shown, the present invention, a method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor, includes the following steps:

[0024] S1. Collect the power curve data of the injection molding machine and perform data preprocessing;

[0025] In this embodiment, the data preprocessing is specifically as follows:

[0026] As Figure 2 shown, use prior knowledge to label the collected power curve of the injection molding machine. According to the working state of the injection molding machine, it is divided into eight types, namely mold closing, mold locking, injection, holding pressure, plasticizing, mold opening, ejector pin ejection, and standby; determine the working state of the injection molding machine at each time point, label the working state of the injection molding machine at each time point of the power curve of the injection molding machine, and then obtain the frequency domain features through Fourier synchronous compression transformation of the time domain power data, and establish a dataset of time-frequency fusion features. The dataset of time-frequency fusion features consists of the time domain and frequency domain of the power data. The time domain signal is expressed as:

[0027]

[0028] where t represents the center position of the window function, k represents the number of components, A k (t) represents the instantaneous amplitude, represents the instantaneous phase;

[0029] The frequency domain signal after Fourier synchronous compression transformation is expressed as:

[0030]

[0031] where T(t,ω) represents the frequency distribution of the power signal after Fourier synchronous compression transformation, g (0) represents the window function at time 0, δ is the Dirac function, S(t,f) is the short-time Fourier transform signal of the power signal S(t), and ω(t,f) is the compressed instantaneous frequency;

[0032] Finally, normalize the data, which is expressed as:

[0033]

[0034] where X is the value of the current data, X min is the minimum value of a row of data, X max is the maximum value of a row of data, X new is the value after normalization.

[0035] S2. Build a BiLSTM-Attenion network and perform iterative training on it;

[0036] In this embodiment, as Figure 3 shown, the BiLSTM-Attenion network specifically includes an input layer, a BiLSTM layer, an attention mechanism layer, and an output layer;

[0037] The input layer, which receives the original data, i.e., the time-frequency fusion features with a dimension of 51;

[0038] The BiLSTM (Bidirectional Long Short-Term Memory Neural Network) layer, which consists of two LSTMs in two directions (the forward LSTM and the backward LSTM). One processes the forward information and the other processes the backward information, and then the results are concatenated;

[0039] The attention mechanism layer, which implements a non-normalized linear transformation based on scores and is then normalized by the softmax function. Specifically:

[0040] First, calculate the attention scores. For each hidden state h t , an unnormalized attention score e is obtained through a linear transformation t :

[0041]

[0042] where is the weight matrix and b a is the bias;

[0043] Then, normalize e t to obtain the attention weight α t :

[0044]

[0045] The obtained weights are used to measure the attention of the model to different parts of the input sequence;

[0046] The output layer, which consists of a fully connected layer and the softmax function, maps the results from the attention mechanism layer to the expected output space and outputs the classification results through the fully connected layer plus the softmax activation function.

[0047] For the BiLSTM-Attenion network, the data with time-frequency fusion features of dimension 51 is input and then trained through the BiLSTM layer. The number of hidden layers of the BiLSTM network is finally optimized to 64 after continuous optimization;

[0048] When the data passes through the BiLSTM layer, for each time step t, there are two hidden states: one from the forward and one from the backward The two hidden states are concatenated together to form a new hidden state This new hidden state is used as the output of the BiLSTM layer and passed to the next layer;

[0049] Finally, after passing through the attention mechanism layer and the fully connected layer, the output is the results of 8 categories, and the one with the highest probability is taken as the final classification result.

[0050] The iterative training of the BiLSTM-Attention network includes:

[0051] Set the number of training epochs, training batches, and training rate, and train the BiLSTM-Attention network. When the network learns the time-frequency features from the injection molding cycle data (collected power data), through iterative training, a BiLSTM-Attention network with a relatively ideal recognition accuracy can be obtained.

[0052] As Figure 4 shown, it is the recognition result diagram of the injection molding process of the trained BiLSTM-Attention network in this embodiment. It can be seen from the figure that the recognition accuracy of each process is very ideal, and the overall recognition accuracy reaches 98.5%.

[0053] S3. Construct a finite state machine based on the injection molding process;

[0054] A finite state machine (FSM, finite state machine) is a mathematical model used to describe the situation of a finite number of states of an object and the transitions and actions between these states. It consists of a set of states, an initial state, inputs, and transition rules based on the inputs and the existing states. The state machine is deterministic and is widely used in computer science and engineering applications. Since the actions of the injection molding cycle of the injection molding machine are fixed, and the basic working process is clamping, mold locking, injection, holding pressure, plasticizing, mold opening, and ejector pin ejection, a finite state machine based on the injection molding process can be constructed.

[0055] In this embodiment, specifically:

[0056] According to the injection molding cycle of the injection molding machine, construct a finite state machine based on the injection molding process, including clamping, mold locking, injection, holding pressure, plasticizing, mold opening, ejector pin ejection, and standby state;

[0057] Program using the Python language according to the state transition relationship, and write a finite state machine based on the injection molding process. The input is the next state of the injection molding machine, and the output is whether to perform a state transition. As Figure 5 shown, it is the schematic diagram of the finite state machine constructed in this embodiment.

[0058] As Figure 5 shown, the specific state transition relationship is:

[0059] Set the initial state as the standby state;

[0060] The condition for the initial state to enter the clamping state is that the injection molding machine changes from standby to clamping and remains in the clamping state for three time steps, each time step being 0.1 s; if not satisfied, the original state is maintained;

[0061] The condition for the clamping state to enter the mold-locking state is: it can only change from the clamping state to the mold-locking state and stay in the mold-locking state for three time steps; if not satisfied, it remains in the original state.

[0062] The condition for transferring to the standby state after mold-locking is: after the injection molding machine completes mold-locking, it transfers to the standby state and stays in the standby state for three time steps; if not satisfied, it remains in the original state, and the previous state before this standby state is saved as mold-locking. This condition is a necessary condition for transferring from the standby state to the injection state.

[0063] The condition for transferring to the injection state is: it can only be transferred from the standby state, and the previous state before standby is the mold-locking state. If it stays in the injection state for three time steps, it enters the injection state; if not satisfied, it remains in the original state.

[0064] The condition for the injection state to enter the pressure-holding state is: it can only be transferred from the injection state to the pressure-holding state and stay in the pressure-holding state for three time steps; if not satisfied, it remains in the original state.

[0065] The condition for the pressure-holding state to enter the plasticizing state is: it can only be transferred from the pressure-holding state to the plasticizing state and stay in this state for three time steps; if not satisfied, it remains in the original state.

[0066] The condition for entering the standby state after plasticizing is completed is: after plasticizing is completed, it transfers to the standby state and stays in the standby state for three time steps; if not satisfied, it remains in the original state, and the previous state before this standby state is saved as plasticizing. This condition is a necessary condition for transferring from the standby state to the mold-opening state.

[0067] The condition for transferring to the mold-opening state is: it can only be transferred from the standby state, and the previous state before standby is the plasticizing state. If it stays in the mold-opening state for three time steps, it enters the mold-opening state; if not satisfied, it remains in the original state.

[0068] The condition for transferring to the standby state after mold-opening is completed is: it stays in the standby state for three time steps; if not satisfied, it remains in the original state.

[0069] The condition for entering the ejector pin ejection state is: it can only be entered from the standby state, and the previous state before standby is the mold-opening state.

[0070] S4. The fused trained BiLSTM-Attenion network and the finite state machine based on the injection molding process identify the injection molding process and generate an identification result curve with process labels; in this embodiment, specifically:

[0071] The fused trained BiLSTM-Attenion network and the finite state machine based on the injection molding process input the process identification result output by the BiLSTM-Attenion network into the finite state machine based on the injection molding process, and obtain an identification result curve with process labels through the determination of the finite state machine.

[0072] As shown Figure 6 in the figure, it is the recognition result of a section of power data passing through the BiLSTM-Attenion network. The orange curve is the active power curve of the motor of the injection molding machine, and the blue curve is the process recognition result. 0-7 respectively represent eight states of the recognized injection molding machine. It can be seen that there will be certain abnormal points during the plasticizing stage, that is, the misdetection situation existing in the recognition of the BiLSTM-Attenion network. The classification result is input into the finite state machine based on the injection molding process, and the incorrect process recognition situation is corrected through the determination of the finite state machine to obtain the corrected recognition result. As shown Figure 7 in the figure, it is the recognition result of the injection molding process after the fusion model. It can be seen from the figure that the recognition result completely conforms to the operating state of the injection molding machine, which is beneficial to the extraction of injection molding process data.

[0073] S5. Extract the process data of each process according to the recognition result curve with process labels; in this embodiment, specifically:

[0074] According to the obtained recognition result curve with process labels, the active power curves of each process are extracted according to different process labels, and then the start and end times, process durations, and operating states of each process are obtained according to the active power curves of each process. These data can provide data resources for the subsequent monitoring of the operating state and fault diagnosis of the injection molding machine.

[0075] The start and end times of the process are obtained according to the time nodes of state transition. The process duration is the holding time of each state, and the operating state of the process is monitored according to the power curve.

[0076] It should also be noted that in this specification, terms such as "including", "comprising" or any other variant thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the said element.

[0077] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor is characterized in that: The following steps are involved: S1. Collect the power curve data of the injection molding machine and perform data preprocessing; S2. Build a BiLSTM-Attenion network and perform iterative training. S3, construct a finite state machine based on the injection molding process; S4, integrating the trained BiLSTM-Attenion network with the finite state machine based on the injection molding process, identifying the injection molding process, and generating a recognition result curve with process labels; S5. Extract process data of each process according to the recognition result curve with process labels.

2. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 1, characterized in that: In step S1, data preprocessing is specifically as follows: The collected power curve of the injection molding machine is annotated by using prior knowledge. It is divided into eight types according to the working status of the injection molding machine, namely mold closing, mold locking, injection, pressure holding, melt melting, mold opening, ejector ejection and standby. The working status of the injection molding machine at each time point is determined, and the working status of the injection molding machine power curve at each time point is annotated. Then, the time domain power data is transformed by Fourier synchronous compression to obtain the frequency domain features, and a data set of time-frequency fusion features is established. Finally, the data is normalized and expressed as: Among them, X is the value of the current data, X min is the minimum value of a row of data, X max is the maximum value of a row of data, X new is the normalized value.

3. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 2, characterized in that: The data set of time-frequency fusion features consists of the time domain and frequency domain of power data. The time domain signal is expressed as: Among them, t represents the center position of the window function, k represents the number of components, and A k (t) represents the instantaneous amplitude, represents the instantaneous phase; The frequency domain signal after Fourier synchronous compression transform is expressed as: Where T(t,ω) represents the frequency distribution of the power signal after Fourier synchronous compression transformation, g (0) represents the window function at time 0, δ is the Dirac function, S(t,f) is the short-time Fourier transform signal of the power signal S(t), and ω(t,f) is the instantaneous frequency after compression.

4. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 1, characterized in that: In step S2, the BiLSTM-Attenion network specifically includes an input layer, a BiLSTM layer, an attention mechanism layer, and an output layer; The input layer receives the original data, i.e., the data of time-frequency fusion features with a dimension of 51; The BiLSTM layer consists of two directional LSTMs, a forward LSTM and a reverse LSTM, one for processing forward information and the other for processing reverse information; The attention mechanism layer implements a non-normalized linear transformation based on the score, which is then normalized by the softmax function. Specifically: First, the attention score is calculated for each hidden state h t , an unnormalized attention score e is obtained through linear transformation t : in, is the weight matrix, b a is bias; Then normalize e t Get the attention weight α t : The obtained weights are used to measure the model's attention to different parts of the input sequence; The output layer, consisting of a fully connected layer and a softmax function, maps the results from the attention mechanism layer to the expected output space and outputs the classification results through a fully connected layer plus a softmax activation function.

5. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 4, characterized in that: For the BiLSTM-Attenion network, the input dimension is 51, and the data of time-frequency fusion features is trained through the BiLSTM layer. The number of hidden layers of the BiLSTM layer is 64. When the data passes through the BiLSTM layer, for each time step t, there are two hidden states: and from the backward LSTM The two hidden states are concatenated together to form a new hidden state The new hidden state is passed to the next layer as the output of the BiLSTM layer; Finally, the attention mechanism layer and the fully connected layer output the results into 8 categories, and the one with the highest probability is taken as the final classification result.

6. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 5, characterized in that: In step S2, the iterative training of the BiLSTM-Attention network includes: Set the training rounds, training batches, and training rate to train the BiLSTM-Attention network. The network learns the time-frequency features from the injection molding cycle data. Through iterative training and learning, a BiLSTM-Attention network with a preset recognition accuracy is obtained.

7. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 1, characterized in that: Step S3 is specifically as follows: According to the injection molding cycle of the injection molding machine, a finite state machine based on the injection molding process is constructed, including mold closing, mold locking, injection, pressure holding, melt melting, mold opening, ejector ejection and standby state; According to the state transition relationship, Python language is used to program a finite state machine based on the injection molding process. The input is the next state of the injection molding machine, and the output is whether to perform state transition.

8. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 7, characterized in that: The specific state transition relationship is: The initial state is set to standby state; The conditions for the initial state to enter the mold closing state are: the injection molding machine switches from standby to mold closing and maintains three time steps in the mold closing state, each time step is 0.1s; if it is not satisfied, the original state is maintained; The conditions for the mold closing state to enter the mold locking state are: it can only be changed from mold closing to mold locking and stay in the mold locking state for three time steps; if it is not satisfied, the original state is maintained; The conditions for transferring to the standby state after mold locking are as follows: the injection molding machine transfers to the standby state after mold locking is completed, and stays in the standby state for three time steps; if it is not satisfied, the original state is maintained, and the state before the current standby state is saved as mold locking. This condition is a necessary condition for transferring from the standby state to the injection state; The conditions for transferring to the injection state are: it can only be transferred from the standby state, and the state before the standby state is the mold locking state, and it enters the injection state after staying in the injection state for three time steps; if it is not satisfied, it will keep the original state; The conditions for the injection state to enter the pressure holding state are: it can only be transferred from the injection state to the pressure holding state, and stay in the pressure holding state for three time steps; if it is not satisfied, the original state is maintained; The conditions for entering the melt state from the holding pressure state are: it can only be transferred to the melt state from the holding pressure state and stay in this state for three time steps; if it is not satisfied, the original state is maintained; The conditions for entering the standby state after the melting of glue is completed are: after the melting of glue is completed, it will enter the standby state and stay in the standby state for three time steps; if it is not satisfied, it will keep the original state and save the state before the standby state as melting of glue. This condition is a necessary condition for switching from the standby state to the mold opening state; The conditions for transferring to the mold opening state are: it can only be transferred from the standby state, and the state before the standby state is melting glue, and it stays in the mold opening state for three time steps before entering the mold opening state; if it is not satisfied, it will keep the original state; The condition for transferring to the standby state after the mold opening is completed is: stay in the standby state for three time steps; if it is not satisfied, keep the original state; The condition for entering the ejector ejection state is: it can only be entered from the standby state, and the state before the standby state is the mold opening state.

9. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 8, characterized in that: Step S4 is specifically as follows: The trained BiLSTM-Attenion network is integrated with the finite state machine based on the injection molding process. The process recognition result output by the BiLSTM-Attenion network is input into the finite state machine based on the injection molding process. After the judgment of the finite state machine, the recognition result curve with process labels is obtained.

10. The method for extracting injection molding process data based on the electrical parameter signal of the oil pump motor according to claim 9, characterized in that: Step S5 is specifically as follows: According to the obtained recognition result curve with process labels, the active power curve of each process is extracted according to different process labels, and then the start and end time, process duration and operation status of each process are obtained according to the active power curve of each process; The start and end time of the process is obtained according to the time node of the state transfer, the process duration is the holding time of each state, and the operating status of the process is obtained by monitoring the power curve.

Images