Injection molding machine screw wear prediction and control method and system based on multi-dimensional fusion

Abstract

The present application relates to a multi-dimensional fusion-based injection molding machine screw wear prediction and control method and system, relating to the field of injection molding machine PLC technology control, which comprises obtaining multi-dimensional sensing signals during the operation of the injection molding machine screw and barrel; analyzing and fusing the multi-dimensional sensing signals to generate a fusion feature vector; arranging the fusion feature vectors of continuous multiple time points within a time length in chronological order to construct an input time sequence; inputting the input time sequence into a CNN-LSTM deep learning model to predict the remaining service life and wear grade; automatically adjusting the key process parameters of the injection molding machine according to the wear grade and grade parameter correction table; if the remaining service life is lower than the life threshold, outputting a maintenance prompt signal; when a maintenance completion signal is received, recording the actual service life of the screw and feeding back to the CNN-LSTM deep learning model for updating. The present application has the effect of enhancing the prediction accuracy of screw wear trend and active intervention ability.

Description

Technical Field

[0001] This invention relates to the field of PLC technology control for injection molding machines, and in particular to a method and system for predicting and controlling screw wear in injection molding machines based on multi-dimensional fusion. Background Technology

[0002] The screw and barrel of an injection molding machine are core actuators in the injection molding process. Operating under high temperature, high pressure, and high shear force conditions for extended periods, they are prone to wear, corrosion, and fatigue damage. Their wear directly affects the plasticization quality of the material, injection accuracy, and the yield rate of the finished product. Sudden failure can not only cause production downtime and losses but also pose safety hazards.

[0003] Currently, the industry mainly divides the maintenance methods for screws and barrels into three categories: reactive maintenance with replacement after failure, periodic maintenance with replacement according to fixed cycles, and condition monitoring based on simple threshold judgment, supplemented by routine maintenance methods such as manual cleaning, parameter adjustment, and visual inspection.

[0004] Regarding the aforementioned technologies, traditional threshold judgment methods mostly rely on data collection from a single sensor, which can only trigger alarms when wear has progressed to a more severe stage, and it is difficult to automatically adjust process parameters to delay component wear. Overall operation and maintenance remains in an open-loop mode of "monitoring-alarm-manual handling," without forming a closed-loop management of the entire process from trend prediction, spare parts preparation, maintenance execution to effect review, and there is still room for improvement. Summary of the Invention

[0005] To enhance the accuracy of screw wear trend prediction and proactive intervention capabilities, this invention provides a method and system for predicting and controlling screw wear in injection molding machines based on multi-dimensional fusion.

[0006] In a first aspect, the present invention provides a method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion, employing the following technical solution: A method for predicting and controlling screw wear in injection molding machines based on multi-dimensional fusion includes: Step 1: Acquire multi-dimensional sensor signals during the operation of the injection molding machine screw and barrel; Step 2: Analyze and fuse the multidimensional sensor signals to generate the fused feature vector for the current moment; Step 3: Arrange the fused feature vectors of multiple consecutive moments within a preset time period in chronological order to construct the input time series; Step 4: Input the input time series into the preset CNN-LSTM deep learning model to predict the remaining service life and wear level; Step 5: Automatically adjust the key process parameters of the injection molding machine according to the wear level and the preset level parameter correction table; Step 6: If the remaining service life is lower than the preset service life threshold, output a preset maintenance prompt signal; Step 7: When the preset maintenance completion signal is received, record the actual service life of the screw and feed it back to the CNN-LSTM deep learning model for updating.

[0007] By adopting the above technical solution, multi-dimensional sensor signals are collected, fused feature vectors are generated, and input time series are constructed. The remaining service life and wear level are predicted by the CNN-LSTM deep learning model. Key process parameters are adjusted according to the wear level and level parameter correction table. When the remaining service life is lower than the service life threshold, maintenance work orders and spare parts purchase requests are generated. After receiving the maintenance completion signal, the actual service life is recorded and the CNN-LSTM deep learning model is updated. This solves the problems of traditional methods, such as single data collection, inability to predict wear levels at each stage, and difficulty in automatically adjusting process parameters to delay component wear. It enhances the prediction accuracy and proactive intervention capability of screw wear trend.

[0008] Optionally, methods for analyzing and fusing multidimensional sensing signals to generate a fused feature vector at the current moment include: Step 20: The multidimensional sensing signals include vibration signals, screw torque signals, barrel temperature signals, screw back pressure signals, and screw current signals; Step 21: Time synchronization of the multidimensional sensor signals to obtain time-aligned multi-channel signals; Step 22: Preprocess and extract features from the multi-channel signals to obtain multi-channel features, including vibration features, screw torque features, barrel temperature features, screw back pressure features, and screw current features; Step 23: Concatenate and reduce the dimensions of the multi-channel features to generate the fused feature vector at the current time.

[0009] By adopting the above technical solution, the multidimensional sensing signal includes vibration signal, screw torque signal, barrel temperature signal, screw back pressure signal and screw current signal. The multidimensional sensing signal is time-synchronized to obtain multi-channel signal. The multi-channel signal is preprocessed and feature extracted to obtain multi-channel feature. The multi-channel feature is spliced ​​and dimensionality reduced to generate the fused feature vector at the current moment. This solves the problem that the existing technology relies on only a single data point, resulting in an incomplete characterization of screw wear state.

[0010] Optional, also includes: Step 24: When a preset cleaning start signal is received, a preset cleaning enhancement mode is used to acquire multi-dimensional sensing signals; Step 25: Obtain the unique cleaning characteristics during the cleaning process; Step 26: Combine the material cleaning features and multi-channel features to generate a fused feature vector.

[0011] By adopting the above technical solution, when a preset cleaning start signal is received, a preset cleaning enhancement mode is used to obtain multi-dimensional sensing signals, and the cleaning features corresponding to the cleaning process are extracted. The cleaning features are then spliced ​​with multi-channel features to generate a fused feature vector, which solves the problem of ignoring the wear impact of the cleaning process on the screw and realizes the special monitoring of wear risk in the cleaning process.

[0012] Optionally, it also includes a material cleaning optimization method when a preset material cleaning start signal is received, the method comprising: Step 240: Obtain the current wear level, the first melting temperature of the material before cleaning, and the second melting temperature of the material after cleaning; Step 241: Select the corresponding cleaning speed curve from the preset speed curve table according to the current wear level; Step 242: Calculate the temperature difference between the first melting temperature and the second melting temperature; Step 243: Calculate the segmented temperature transition curve based on the temperature difference and the preset temperature change rate; Step 244: Retrieve the cleaning time coefficient corresponding to the current wear level and multiply it by the preset basic cleaning time to calculate the suggested time value; Step 245: Send the material cleaning speed curve, segmented temperature transition curve, and suggested time value to the preset PLC controller and execute.

[0013] By adopting the above technical solution, after triggering the material clearing start signal, the current wear level, first melting temperature and second melting temperature are obtained, the material clearing speed curve is matched, and a segmented temperature transition curve is generated based on the temperature difference and temperature change rate. The suggested time value is calculated by combining the material clearing time coefficient and the basic material clearing time, and then sent to the PLC controller to execute the material clearing operation. This solves the problem that existing injection molding machines lack adaptive optimization during the material clearing process and are prone to accelerated screw wear due to sudden temperature changes and improper speed, effectively extending the screw service life.

[0014] Optionally, methods for automatically adjusting key process parameters of the injection molding machine based on wear level and a preset level parameter correction table also include: Step 50: When the wear level is not the preset health level, find the correction amount corresponding to the wear level according to the wear level and level parameter correction table; Step 51: Obtain the current process parameters, and calculate the corrected key process parameters based on the correction amount. Step 52: Send the key process parameters to the preset PLC controller to adjust the screw speed, back pressure, barrel temperature and injection parameters.

[0015] By adopting the above technical solution, when the wear level is not the preset health level, the corresponding correction amount is queried according to the wear level and level parameter correction table, the current process parameters are obtained, and the corrected key process parameters are calculated by combining the correction amount. The key process parameters are sent to the PLC controller to realize the adjustment of screw speed, back pressure, barrel temperature and injection parameters. This solves the problem that after the screw wears, it can only be passively alarmed and cannot be actively intervened to delay the wear. It achieves the effect of adaptive process parameter adjustment according to the wear level.

[0016] Optionally, it also includes a safety constraint method for automatically adjusting key process parameters of the injection molding machine, the method comprising: Step 53: When the critical process parameter falls within the preset safety range, the current process parameter is gradually adjusted to the critical process parameter. Step 54: Continuously obtain product quality indicators; Step 55: If the product quality index exceeds the preset allowable range, revert the key process parameters to the current process parameters before adjustment and output the preset adjustment alarm signal.

[0017] By adopting the above technical solution, when the key process parameters fall within the safe range, the parameter adjustment is completed in a gradual manner; the product quality indicators are continuously acquired, and if the product quality indicators exceed the allowable range, the key process parameters are reverted to the current process parameters before the adjustment, and an adjustment alarm signal is output, which solves the problem of product quality fluctuation caused by parameter mutations or over-adjustment in adaptive adjustment.

[0018] Optional, also includes: Step 56: If the wear level obtained by continuous prediction is the preset alarm level, count the number of consecutive alarms when the wear level is the alarm level. Step 57: If the number of consecutive alarms exceeds the preset number of alarms and the remaining service life is lower than the preset shutdown threshold, trigger the preset shutdown protection signal.

[0019] By adopting the above technical solution, the number of consecutive alarms is counted when the alarm level is continuously predicted. When the number of consecutive alarms is greater than the number of alarms and the remaining service life is lower than the shutdown threshold, a shutdown protection signal is triggered, thus avoiding the problem of occasional false alarms affecting normal production.

[0020] Optionally, if the remaining service life is lower than a preset service life threshold, the method for outputting a preset maintenance prompt signal includes: Step 60: Calculate the latest maintenance time based on the remaining service life and the preset safety margin; Step 61: Obtain the production schedule plan, find the maintainable time window in the production schedule plan based on the latest maintenance time, and generate a maintenance work order; Step 62: Find the required quantity of spare parts based on the maintenance work order and check the current inventory. Step 63: If the current inventory is less than the spare parts requirement, generate a spare parts purchase request and output a maintenance prompt signal.

[0021] By adopting the above technical solution, when the remaining service life is lower than the service life threshold, the latest maintenance time is calculated by combining the remaining service life and safety margin. The maintenance time window is determined by combining the production scheduling plan and a maintenance work order is generated. The spare parts demand and current inventory are checked. If the inventory is insufficient, a spare parts purchase request is generated. This solves the problem of lack of proactive maintenance scheduling and spare parts management in the existing technology and realizes closed-loop management of the whole process from prediction to maintenance execution to spare parts preparation.

[0022] Optionally, when a preset maintenance completion signal is received, methods for recording the actual service life of the screw and feeding it back to the CNN-LSTM deep learning model for updating include: Step 70: Define the final remaining useful life as the last predicted useful life before maintenance; Step 71: Compare the actual service life with the final remaining service life and calculate the prediction deviation; Step 72: Feed the prediction bias back to the CNN-LSTM deep learning model to update the CNN-LSTM deep learning model.

[0023] By adopting the above technical solution, the maintenance completion signal is received, the final remaining lifespan is determined, the prediction deviation is calculated in combination with the actual lifespan, and the feedback is sent to the CNN-LSTM deep learning model to complete the update. This solves the problem in the existing technology that it is impossible to optimize the model using actual operating data, and realizes a complete closed loop from actual operating data to model optimization.

[0024] Secondly, this invention provides a multi-dimensional fusion-based injection molding machine screw wear prediction and control system, employing the following technical solution: A multi-dimensional fusion-based injection molding machine screw wear prediction and control system includes: The acquisition module is used to acquire multi-dimensional sensor signals, material cleaning characteristics, current process parameters, product quality indicators, and production scheduling plans; The memory is used to store the program of the injection molding machine screw wear prediction and control method based on multi-dimensional fusion as described above; The processor loads and executes programs from memory.

[0025] In summary, the present invention has at least one of the following beneficial technical effects: It solves the problems of traditional methods in collecting data in a single way, being unable to predict the wear level at each stage, and being unable to automatically adjust process parameters to delay component wear. It achieves closed-loop management of the entire process from trend prediction, spare parts preparation, maintenance execution to effect review. It solves the problems of lack of adaptive optimization in the material cleaning process of existing injection molding machines, and the easy acceleration of screw wear due to sudden temperature changes and improper speed, effectively extending the service life of the screw; This solves the problem that screw wear can only be passively alarmed and cannot be actively intervened to delay wear, and achieves the effect of adaptive adjustment of process parameters according to the wear level. Attached Figure Description

[0026] Figure 1 This is a flowchart of the injection molding machine screw wear prediction and control method based on multi-dimensional fusion in the embodiments of this application; Figure 2 This is a schematic diagram of the overall hardware architecture of the system in another embodiment; Figure 3 This is a flowchart of the material cleaning optimization method in the material cleaning process of this application embodiment. Detailed Implementation

[0027] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments.

[0028] This invention discloses a method for predicting and controlling screw wear in injection molding machines based on multi-dimensional fusion. (Refer to...) Figure 1 The multi-dimensional fusion-based method for predicting and controlling screw wear in injection molding machines includes: Step 1: Acquire multi-dimensional sensor signals during the operation of the injection molding machine screw and barrel.

[0029] Multidimensional sensing signals refer to multiple physical quantity signals reflecting the operating status of the screw and barrel of an injection molding machine. These signals are synchronously acquired by a group of multidimensional sensors installed at key locations on the screw and barrel of the injection molding machine. These signals include vibration signals, screw torque signals, barrel temperature signals, screw back pressure signals, and screw current signals.

[0030] Vibration signals are collected by installing triaxial accelerometers at the screw drive end bearing housing, the front flange of the barrel, and the outer wall of the middle section of the barrel. These sensors have a sampling frequency of ≥10kHz and acquire radial, axial, and tangential vibration signals during screw rotation. Because increased clearance due to screw wear causes changes in the vibration spectrum characteristics, such as increased characteristic frequency amplitude and the appearance of new frequency components, vibration signals are a sensitive indicator for early wear detection.

[0031] The screw torque signal is acquired in real time by installing a torque sensor at the coupling between the screw drive shaft and the reduction gearbox. The measurement accuracy of the torque sensor is ±0.5%FS. Changes in the clearance caused by screw / barrel wear alter the shear resistance of the material, thus causing changes in the torque fluctuation characteristics.

[0032] The barrel temperature signal is acquired by installing thermocouples along the barrel's axial direction at the feeding section, compression section, metering section, and nozzle. The thermocouple measurement accuracy is ±1℃. The final output shows the temperature and temperature gradient distribution of each section of the barrel. Changes in the screw / barrel clearance due to wear can affect the material's shear heat generation and heat conduction characteristics, causing abnormal changes in the temperature distribution.

[0033] The screw back pressure signal is acquired in real time by installing a back pressure sensor at the bottom of the screw. The accuracy of the back pressure sensor is ±0.25%FS. Because the increased backflow caused by wear in the screw metering section will cause changes in the back pressure fluctuation characteristics, back pressure stability is an important indicator of the wear state of the screw metering section.

[0034] The screw current signal is obtained by installing a current sensor in the screw drive motor power supply circuit to acquire the drive motor current signal, i.e., the screw current signal. Because the current fluctuation characteristics are directly related to the screw load changes, it can be used as an auxiliary criterion for judging the wear state.

[0035] Step 2: Analyze and fuse the multidimensional sensor signals to generate the fused feature vector at the current moment.

[0036] The fused feature vector refers to the comprehensive feature data of the screw's current wear state obtained after time synchronization, feature extraction, stitching, and dimensionality reduction processing of multi-dimensional sensor signals. The method for obtaining the fused feature vector is explained in subsequent steps.

[0037] Step 3: Arrange the fused feature vectors of multiple consecutive moments within a preset time period in chronological order to construct the input time series.

[0038] The time span refers to the time range covered by the fusion feature vectors selected when constructing the input time series, which is predetermined by those skilled in the art based on the evolution cycle of screw wear and model training experience. For example, the time span can be set to 7 to 14 days. During the operation of the injection molding machine, the system collects and stores fusion feature vectors at fixed sampling intervals, such as once every hour. When it is necessary to construct the input time series, all fusion feature vectors within the most recent time span are retrieved from the stored historical data.

[0039] The temporal order refers to the order in which the fused feature vectors are arranged according to the sequence of their acquisition times. Each fused feature vector is associated with a corresponding timestamp during generation, provided by the system clock. When constructing the input time series, the system acquires the fused feature vectors within a specified time length and arranges them in ascending order of timestamps, so that the vectors in the sequence correspond to times from earliest to latest.

[0040] Input time series refers to sequence data formed by arranging fused feature vectors from multiple consecutive moments in chronological order. The method for acquiring input time series is as follows: fused feature vectors at fixed time intervals are collected and stored, for example, once every hour; multiple consecutive fused feature vectors within the most recent preset time length are selected, for example, fused feature vectors from the most recent 7 to 14 days; these are arranged in chronological order to form the input time series; if the sequence length is less than the preset length, it is padded with zeros or by repeating the first frame.

[0041] Step 4: Input the input time series into the preset CNN-LSTM deep learning model to predict the remaining lifespan and wear level.

[0042] The CNN-LSTM deep learning model refers to a deep learning model that employs a dual-channel fusion architecture of convolutional neural networks and long short-term memory networks. This model is integrated into an edge computing unit deployed locally on the injection molding machine. This edge computing unit can be integrated into a PLC control cabinet or deployed independently. It is responsible for real-time inference calculations on the multi-dimensional feature vectors output from the data acquisition module, outputting screw / barrel wear status assessment, remaining service life (RUL) prediction, and wear trend warning information. The edge computing unit uses an industrial-grade embedded computing platform with GPU / NPU acceleration capabilities, supporting local model inference to ensure real-time prediction results with an inference latency of no more than 500ms. Furthermore, production data does not need to be uploaded to the cloud, ensuring data security.

[0043] The CNN-LSTM deep learning model includes a CNN local feature extraction channel, an LSTM long-term trend learning channel, and a fusion decision layer.

[0044] 1. CNN Channel The input time series data is fed into the CNN channels. The input data contains multiple consecutive time points, each corresponding to a D-dimensional fused feature vector. It is processed through three layers of one-dimensional convolutions, with kernel sizes of 7, 5, and 3, and channel numbers of 64, 128, and 256 respectively. Each layer sequentially performs convolution, batch normalization, ReLU activation, and max pooling to obtain a feature sequence of length L, where L is the length in the time dimension after convolution and pooling. This represents the feature value at the i-th time step in the sequence; finally, global average pooling is used, which involves summing the feature values ​​of all L time steps and dividing by L to output the feature vector. .

[0045]

[0046] 2. LSTM Channel Given the same time series, long-term degradation trends are learned through 2 to 3 stacked LSTM layers. Each LSTM unit updates its memory state and hidden state through a triple gating mechanism of forget gate, input gate, and output gate. The core update equation is: Among them, C t C represents the current memory state. t-1 h represents the memory state of the previous moment. t-1 x is the hidden state from the previous time step. t For the current input, W f W i W C Let b be the weight matrix. f b i b C For bias terms, is the sigmoid activation function, and tanh is the hyperbolic tangent activation function. This indicates element-wise multiplication.

[0047] The final layer output is weighted and aggregated using an attention mechanism, adaptively focusing on the time period that is most discriminative for wear prediction: , in, The feature vector is the output of the LSTM channels after weighted aggregation through the attention mechanism, where T is the time step of the input sequence, and h is the number of time steps. t Let be the hidden state output by the LSTM at time t. Let be the attention weight at time t. This is the weight matrix. For bias terms, denoted by , where exp is the exponential function, tanh is the hyperbolic tangent activation function, ∑ is the summation symbol, and the denominator is the sum of attention scores at all time steps, used to normalize the attention weights.

[0048] 3. Integrating the decision-making level The two-channel feature vectors are concatenated into V_fusion=[V_cnn;V_lstm], and after passing through two fully connected layers, they are divided into two output heads: RUL regression and wear level classification. A multi-task learning strategy is used for joint training. .

[0049] Here This refers to the RUL value predicted by the model, where yᵢ is the actual remaining lifetime label. The first part, the MSE loss, is used to train the model to accurately predict the RUL. y is the total loss value for multi-task learning, N is the number of samples, and y is the total loss value for multi-task learning. i Let y be the actual remaining useful life of the i-th sample, and ŷᵢ be the predicted remaining useful life of the i-th sample. i -ŷᵢ) 2 Mean squared error loss, used for RUL regression tasks; y ik Let y be the true label of the i-th sample belonging to the k-th wear level. ik P takes the value 0 or 1. ik Let be the predicted probability that the i-th sample belongs to the k-th wear level. Cross-entropy loss is used for wear level classification tasks, with a total of 4 levels; α and β are the weight coefficients of regression loss and classification loss, respectively.

[0050] Among them, the actual value of remaining service life y i This refers to the cumulative running time remaining from the time of the i-th sample collection to the final replacement time of the screw. It is obtained as follows: the PLC controller continuously records the cumulative running time of the screw and simultaneously records the running time at each sample collection time. When the screw meets the replacement conditions and is replaced, the PLC controller records the cumulative running time of the screw from its initial commissioning to the replacement time, which is taken as the actual service life of the screw. During the historical data processing phase, the full lifecycle running data is divided into multiple samples according to preset time intervals. For each sample, the actual service life of the screw is subtracted from the running time at the sample collection time to obtain the true value of the remaining service life of that sample. It should be noted that the above cumulative running time is based on the actual rotation and running time of the screw, excluding non-running periods such as equipment shutdown, standby, and mold changing, to ensure that the lifespan data has a consistent statistical benchmark regardless of whether the equipment is in continuous or intermittent operation. The PLC controller is a commonly used injection molding machine control unit in this field, which will be further explained in the embodiment of step 6 below.

[0051] In summary, the actual service life is the total cumulative value of the old screw from commissioning to replacement, which is directly read by the PLC upon completion of maintenance. The remaining service life, on the other hand, is the cumulative remaining runtime of the screw at a specific historical sampling point. It is calculated by combining the runtime recorded by the PLC at that moment with the actual service life read upon completion of maintenance; that is, the actual service life minus the runtime already run. Numerically, they are related as "total" and "component," and in application, they correspond to the feedback label during model updates and the supervision label during the training phase, respectively.

[0052] The CNN-LSTM deep learning model is trained using supervised learning data covering its entire lifecycle, including normal operation, wear and tear stages, and maintenance / replacement records. The Adam optimizer is used with an initial learning rate of 1×10⁻⁶. -3 A cosine annealing strategy is used for dynamic decay. The training and validation sets are divided in an 8:2 ratio. An early stopping mechanism based on the validation set loss is used to prevent overfitting. This early stopping mechanism has a waiting period of 10 rounds, and the parameter update rule is as follows: θ t Let θ be the model parameters at step t. t+1 Here are the updated model parameters, where η is the learning rate, initially set to 1×10. -3 m t and v t These are the first-order moment estimate and the second-order moment estimate of the gradient, respectively. and These are the first-order moment estimates and second-order moment estimates after bias correction, β1 t and β2 t β1 and β2 raised to the power of t, respectively, are used for bias correction. This is a smoothing term used to prevent division by zero errors.

[0053] The edge computing unit triggers inference every hour, inputting the fused feature sequence from the most recent T days into a trained CNN-LSTM model, where T is 7 to 14 days and this value is configurable. It also outputs the RUL prediction value and the wear level probability distribution. The warning determination rule is to take the level with the highest probability in the classification output as the current wear state. Where Grade represents the wear level output, k is the category index of the wear level, and k=1, 2, 3, 4 correspond to the health level, attention level, warning level, and alarm level, respectively. P(class=k) is the probability that the current sample belongs to the kth wear level as predicted by the model classification output head, and argmax indicates that the category index that maximizes the probability value is taken as the output result.

[0054] Remaining service life refers to the remaining operational time of the screw from the current moment until it is expected to fail or need replacement. The input time series is fed into a trained CNN-LSTM deep learning model, and the model's regression output head directly outputs the predicted remaining service life value.

[0055] Wear level refers to the graded assessment of the current wear degree of the screw. The input time series is fed into a trained CNN-LSTM deep learning model, and the model's classification output head outputs the probability distribution of each wear level. The level with the highest probability is taken as the current wear level. Wear levels include health level, attention level, warning level, and alarm level.

[0056] Step 5: Automatically adjust the key process parameters of the injection molding machine according to the wear level and the preset level parameter correction table.

[0057] The wear level correction table contains a mapping relationship between wear levels and process parameter corrections, which are pre-set by those skilled in the art based on experience. First, correction coefficients or correction amounts for each process parameter under different wear levels are defined. These process parameters include screw speed, back pressure, barrel temperatures at various stages, injection speed, and holding pressure. For example, when the wear level is "Caution" level, the screw speed correction coefficient is 0.95, representing a 5% reduction; when the wear level is "Warning" level, the screw speed correction coefficient is 0.85, representing a 15% reduction, the back pressure correction is +3 bar, and the compression section temperature correction is -3°C. The parameter values ​​in this table can be adjusted by engineers on the HMI based on practical experience. The HMI is a human-machine interface integrated into the injection molding machine's control panel, providing a visual display of the screw / barrel wear status, including but not limited to: real-time waveforms and trend charts of multi-dimensional sensor data; wear level; predicted remaining service life; adaptive parameter adjustment records and effect comparisons; maintenance work order status and maintenance scheduling information.

[0058] Key process parameters refer to the core control parameters that affect the plasticizing quality and screw wear rate of the injection molding machine, including screw speed, back pressure, barrel temperature, and injection parameters. The methods for obtaining key process parameters are explained in subsequent steps.

[0059] Step 6: If the remaining service life is lower than the preset service life threshold, output a preset maintenance prompt signal.

[0060] The lifespan threshold refers to the critical remaining service life value that triggers maintenance scheduling and spare parts procurement requests. It is determined by those skilled in the art based on the expected service life of the screw, safety margin, and production scheduling requirements. For example, the lifespan threshold can be set to 336 hours, or 14 days. This threshold can be adjusted by the operator through a human-machine interface.

[0061] A maintenance prompt signal is a control signal that prompts operators or the management system to prepare for screw maintenance. It is pre-set by those skilled in the art. This signal includes the following information: the predicted remaining service life, the suggested maintenance time window, and spare parts requirement information. After generation, the maintenance prompt signal is sent via the communication module to the human-machine interface for display, simultaneously to the manufacturing execution system to trigger a maintenance work order process, and to the enterprise resource planning system to trigger a spare parts inventory query and purchase request process. After viewing the maintenance prompt signal through the human-machine interface, operators can arrange a maintenance plan based on the signal content.

[0062] If the remaining service life is lower than the preset service life threshold, it indicates that maintenance and spare parts preparation are required, and therefore a preset maintenance prompt signal is output.

[0063] Step 7: When the preset maintenance completion signal is received, record the actual service life of the screw and feed it back to the CNN-LSTM deep learning model for updating.

[0064] The maintenance completion signal is a trigger signal indicating that screw replacement or maintenance work has been completed. It is pre-set by those skilled in the art and then generated after confirmation by the operator through the human-machine interface. When the maintenance personnel complete the screw replacement or maintenance operation and confirm the maintenance completion on the injection molding machine's human-machine interface, the system receives the maintenance completion signal.

[0065] Actual service life refers to the cumulative operating time of the old screw from its initial commissioning to replacement. The PLC controller continuously records the cumulative operating time of the currently running screw. When the operator confirms the completion of maintenance through the human-machine interface, the system obtains the cumulative operating time of the screw from commissioning to replacement, which is taken as the actual service life of the screw. This data is used to compare with the previously predicted remaining service life to calculate the prediction deviation and for model updates. The PLC controller is the core control unit of the entire system, responsible for the conventional process control of the injection molding machine and the adaptive parameter control function newly added in this invention. Conventional process control includes injection, holding pressure, cooling, mold opening, and mold closing. The PLC controller is connected to the data acquisition module, IO module group, servo driver, and temperature control module through the EtherCAT bus, and receives the wear prediction results and parameter adjustment instructions output by the edge computing unit in real time, and executes the corresponding process parameter corrections.

[0066] Upon receiving a pre-set maintenance completion signal, it indicates that the old screw has been replaced and the new screw has been installed. Therefore, the cumulative operating time of the old screw from its initial commissioning to the replacement time is read from the PLC controller and taken as the actual service life of the old screw. This actual service life is fed into the CNN-LSTM deep learning model. The model compares this actual service life with the remaining service life predicted for the old screw before maintenance, calculating the prediction deviation. This prediction deviation is used as incremental training data to update the model parameters, improving the accuracy of subsequent predictions.

[0067] Reference Figure 2 , Figure 2This is a schematic diagram of the overall hardware architecture of the system in another embodiment, which includes a multi-dimensional sensor group, a data acquisition and preprocessing module, a PLC controller and edge computing unit, an I / O module group and actuators, an HMI (Human-Machine Interface), a communication module, a cloud server, and a MES / ERP system. The multi-dimensional sensor group is installed at key locations on the screw and barrel of the injection molding machine to collect multi-dimensional sensor signals such as vibration, torque, temperature, back pressure, and current in real time. Each sensor signal is led to the data acquisition and preprocessing module via shielded cables. The data acquisition and preprocessing module is installed in the PLC control cabinet and connected to the PLC controller via an EtherCAT bus. This module is responsible for the synchronous acquisition and preprocessing of the multi-dimensional sensor signals. The PLC controller is responsible for the conventional process control and adaptive parameter adjustment of the injection molding machine. The edge computing unit, equipped with a CNN-LSTM model, performs real-time inference on the fused features and outputs the remaining service life and wear level. The O module group connects field switch sensors and actuators. The actuators receive commands from the PLC controller to drive the screw motor, heating coil, and hydraulic cylinder to perform actions. The HMI (Human Machine Interface) is integrated into the injection molding machine's control panel to display wear status, remaining service life, adaptive adjustment records, and maintenance work order information. The communication module is integrated into the PLC control cabinet and supports Ethernet, Wi-Fi, 4G, and 5G communication methods. The cloud server is deployed in a remote computer room and is responsible for the initial and incremental training of the CNN-LSTM model. The MES (Manufacturing Execution System) receives maintenance work orders and coordinates with the production scheduler, while the ERP (Enterprise Resource Planning) system receives spare parts purchase requests and executes the procurement process.

[0068] The methods for analyzing and fusing multidimensional sensor signals to generate a fused feature vector at the current moment include: Step 20: The multidimensional sensing signals include vibration signals, screw torque signals, barrel temperature signals, screw back pressure signals, and screw current signals.

[0069] Step 21: Time synchronization of the multidimensional sensor signals to obtain time-aligned multi-channel signals.

[0070] Multi-channel signals refer to signal datasets where each sensor channel corresponds to the others at the same time after time synchronization processing. A unified sampling clock is used to synchronously trigger acquisition across all sensor channels. For channels with different sampling frequencies, the highest sampling frequency is used as the reference for interpolation and resampling of other channel signals, ensuring that all channel data are aligned at the same time point. Linear interpolation or spline interpolation is used as the interpolation method. The multi-channel signal is obtained after processing using the above time synchronization method.

[0071] Step 22: Preprocess and extract features from the multi-channel signals to obtain multi-channel features, including vibration features, screw torque features, barrel temperature features, screw back pressure features, and screw current features.

[0072] Multi-channel features refer to the set of feature parameters extracted from multi-channel signals. First, the multi-channel signals undergo preprocessing, including denoising, filtering, and outlier removal. Specifically, bandpass filtering is applied to the vibration signal to remove low-frequency base vibration and high-frequency electromagnetic interference; moving average filtering is applied to the barrel temperature signal to eliminate thermocouple measurement noise; and outlier removal is performed on the screw torque and screw back pressure signals. Then, time-domain, frequency-domain, and time-frequency-domain features are extracted from the preprocessed signals to form a multi-dimensional feature vector. Time-domain features include mean, variance, peak value, peak-to-peak value, root mean square value, kurtosis, and skewness; frequency-domain features include dominant frequency amplitude, spectral centroid, and band power ratio. Finally, the multi-dimensional feature vector is normalized, for example using Min-Max normalization or Z-Score standardization, to eliminate the influence of differences in physical dimensions and magnitudes, generating the final multi-channel features.

[0073] Step 23: Concatenate and reduce the dimensions of the multi-channel features to generate the fused feature vector at the current time.

[0074] The feature parameters from the multi-channel features are concatenated in a preset order to form a one-dimensional feature vector. An example concatenation order is: vibration feature - screw torque feature - barrel temperature feature - screw back pressure feature - screw current feature. The dimension of the concatenated feature vector is equal to the sum of the number of feature parameters in each channel. Principal component analysis is then performed on the concatenated feature vector to reduce its dimensionality, removing redundant features and retaining key feature components strongly correlated with wear conditions, resulting in a dimensionality-reduced fused feature vector.

[0075] This also includes: In the normal working mode, the five types of sensor signals are time-synchronized, feature-extracted, and spliced ​​and dimensionality-reduced according to steps 20 to 23 to generate a fused feature vector; during the cleaning process, the system switches to the cleaning working mode, and the cleaning features are extracted according to steps 24 to 26 and spliced ​​with the multi-channel features to generate a fused feature vector. After the cleaning is completed, the normal mode is restored.

[0076] Step 24: When a preset cleaning start signal is received, a preset cleaning enhancement mode is used to acquire multi-dimensional sensing signals.

[0077] The material clearing start signal is a signal that indicates the material clearing process has begun, and it is preset by someone skilled in the art. The material clearing start signal is generated when the operator presses the material clearing start button through the injection molding machine's human-machine interface.

[0078] Enhanced cleaning mode refers to a special operating mode adopted during the cleaning process to increase data acquisition density and capture transient characteristics. Upon receiving a cleaning start signal, the system automatically switches from the normal operating mode to enhanced cleaning mode. In enhanced cleaning mode, the acquisition frequency of multi-dimensional sensor signals is increased to 2 to 5 times that of the normal mode. The specific acquisition frequency is determined by those skilled in the art based on the current wear level and the type of material being cleaned; the higher the wear level or the greater the viscosity change of the material, the greater the increase in acquisition frequency. Enhanced cleaning mode continues until the cleaning process is completed, after which the system automatically reverts to the normal acquisition mode.

[0079] When a preset cleaning start signal is received, it indicates that the injection molding machine is about to enter a material switching or shutdown cleaning operation. At this time, the system automatically switches from the normal acquisition mode to the cleaning enhancement mode, increasing the acquisition frequency of multi-dimensional sensor signals to 2 to 5 times that of the normal mode, and marking the current period as the cleaning process in the data record, in preparation for extracting the unique feature parameters of cleaning. The cleaning enhancement mode continues throughout the entire cleaning process, and automatically returns to the normal acquisition mode after the cleaning is completed.

[0080] Step 25: Obtain the unique cleaning characteristics during the cleaning process.

[0081] Cleaning characteristics refer to specific parameters reflecting the thermal and mechanical shocks experienced by the screw and barrel during the cleaning process. These include temperature change rate, peak torque impact, back pressure difference before and after cleaning, and other optional cleaning characteristics. These cleaning characteristics are extracted from multi-dimensional sensor signals acquired in enhanced cleaning mode: Temperature change rate refers to the rate of temperature change in each section of the barrel during the cleaning process; temperature values ​​of each section of the barrel are continuously acquired, and the temperature change per unit time is calculated, i.e., the temperature difference between the current moment and the previous moment is calculated and then divided by the sampling interval. Peak torque impact refers to the instantaneous maximum value of the screw torque signal during the cleaning process; the screw torque signal is acquired at a sampling frequency of not less than 1kHz, and the maximum amplitude of the torque signal is detected. Back pressure difference before and after cleaning refers to the difference between the back pressure value after the cleaning process ends and the back pressure value before cleaning begins. Other optional cleaning characteristics include, but are not limited to, the main frequency offset of the vibration signal during cleaning, the fluctuation amplitude of the screw current, and the total cleaning time.

[0082] Step 26: Combine the material cleaning features and multi-channel features to generate a fused feature vector.

[0083] In the enhanced cleaning mode, multi-channel features and cleaning features are acquired simultaneously. The multi-channel features include vibration features, torque features, temperature features, back pressure features, and current features. The cleaning features include temperature change rate, torque impact peak value, and back pressure difference before and after cleaning. The cleaning features and multi-channel features are concatenated in a preset order. Based on the multi-channel feature vector, the cleaning features are concatenated to its end in the following order: temperature change rate, torque impact peak value, back pressure difference before and after cleaning, and other optional cleaning features. The temperature change rate includes the change rates at the feeding section, compression section, and metering section. After concatenation, an extended fusion feature vector containing both conventional and cleaning features is formed, which is then input into the deep learning model after dimensionality reduction using principal component analysis.

[0084] Reference Figure 3 It also includes a material cleaning optimization method when a preset material cleaning start signal is received, the method comprising: Step 240: Obtain the current wear level, the first melting temperature of the material before cleaning, and the second melting temperature of the material after cleaning.

[0085] The current wear level refers to the graded assessment result of the current wear degree of the screw. The current input time series is fed into the trained CNN-LSTM deep learning model, and the model's classification output head outputs the probability distribution of each wear level. The level with the highest probability is taken as the current wear level.

[0086] The first melt temperature refers to the temperature at which the material reaches a molten state in the injection molding machine barrel before cleaning. The first melt temperature is selected by the operator through the human-machine interface according to the material grade. This temperature value is equal to the standard processing temperature of the material before cleaning and can be obtained from the material's process parameter table.

[0087] The second melting temperature refers to the temperature at which the material reaches a molten state in the injection molding machine barrel after cleaning. The second melting temperature is selected by the operator through the human-machine interface based on the material grade. This temperature value is equal to the standard processing temperature of the material after cleaning and can be obtained from the material's process parameter table.

[0088] Step 241: Select the corresponding cleaning speed curve from the preset speed curve table according to the current wear level.

[0089] The speed curve table stores the mapping relationship between different wear levels and their corresponding cleaning speed curve parameters. Each curve contains the speed setpoint or speed control parameter for each stage of the cleaning process. The speed curve table is pre-calibrated by the equipment manufacturer according to the injection molding machine model and screw specifications. Operators can view and adjust the parameters in the table through the human-machine interface. Each row in the table corresponds to a wear level, including health level, caution level, warning level, and alarm level. Each column contains the speed setpoint or speed control parameter for each stage of the cleaning process. The higher the wear level, the greater the reduction in cleaning speed relative to the normal cleaning speed. Taking the caution level as an example, the speed curve parameters stored in the speed curve table are: initial speed 60 rpm, target speed 50 rpm, 3 gradual change cycles, and the speed reduction within each cycle is proportionally distributed. The speed curve corresponding to the health level is the normal cleaning speed, without any reduction processing; the warning level and alarm level correspond to lower speed setpoints.

[0090] The cleaning speed curve is a preset trajectory curve showing the change of screw speed over time during the cleaning process. The cleaning speed curve is obtained from the speed curve table based on the current wear level.

[0091] Step 242: Calculate the temperature difference between the first melting temperature and the second melting temperature.

[0092] The temperature difference refers to the absolute difference between the first melting temperature and the second melting temperature. The difference is obtained by subtracting the second melting temperature from the first melting temperature, and then taking the absolute value of this difference.

[0093] Step 243: Calculate the segmented temperature transition curve based on the temperature difference and the preset temperature change rate.

[0094] The temperature change rate refers to the rate at which the barrel temperature changes during the cleaning process. The temperature change rate is predetermined by those skilled in the art based on the thermal shock resistance of the barrel material and the temperature sensitivity of the material. For example, the temperature change rate can be set to 5 degrees Celsius per minute. This rate value can be adjusted by the operator through a human-machine interface.

[0095] A segmented temperature transition curve refers to a temperature change trajectory formed by breaking down the total temperature change during the material cleaning process into multiple stages, with each stage gradually adjusting the temperature at a preset temperature change rate. The temperature difference is segmented using a preset step size, the number of segments is rounded up, and the time required for each segment is calculated using the preset temperature change rate. Starting from the first melting temperature, the temperature is adjusted segment by segment to the second melting temperature, generating a segmented temperature transition curve. For example, each step size can be set to 30 degrees Celsius, and the temperature change rate can be set to 5 degrees Celsius per minute. Taking a first melting temperature of 250 degrees Celsius and a second melting temperature of 190 degrees Celsius as an example, the temperature difference is 60 degrees Celsius, the number of segments is 2, the time for each segment is 6 minutes, and the total transition time is 12 minutes.

[0096] Step 244: Retrieve the cleaning time coefficient corresponding to the current wear level and multiply it by the preset basic cleaning time to calculate the suggested time value.

[0097] The cleaning time coefficient is a factor used to adjust the cleaning time. The cleaning time coefficient is obtained from a preset coefficient table based on the current wear level. This table contains a mapping relationship between different wear levels and their corresponding coefficient values, preset by those skilled in the art. The coefficient for a healthy level is 1.0; for a caution level, it is 1.2; for a warning level, it is 1.5; and for an alarm level, it is 2.0. The higher the wear level, the larger the cleaning time coefficient, indicating a longer cleaning time is required to ensure effective cleaning. This coefficient table can be adjusted by the operator through a human-machine interface.

[0098] The basic cleaning time refers to the minimum time required to complete cleaning under standard operating conditions, which are healthy wear levels and use conventional materials. The basic cleaning time is predetermined by those skilled in the art based on the injection molding machine model, screw specifications, and type of conventional material. For example, the basic cleaning time can be set to 8 minutes. This time value can be adjusted by the operator based on practical experience through the human-machine interface.

[0099] The recommended cleaning time is the suggested cleaning time obtained by adjusting the basic cleaning time according to the current wear level. The recommended time is the product of the found cleaning time coefficient and the basic cleaning time. For example, with a basic cleaning time of 8 minutes and a care level coefficient of 1.2, the recommended time is 9.6 minutes.

[0100] Step 245: Send the material cleaning speed curve, segmented temperature transition curve, and suggested time value to the preset PLC controller and execute.

[0101] Referring to the description in step 7, the PLC controller receives the material cleaning speed curve, segmented temperature transition curve, and suggested time value, and executes the material cleaning process according to these parameters.

[0102] The method for automatically adjusting the key process parameters of the injection molding machine based on the wear level and a preset level parameter correction table also includes: Step 50: When the wear level is not the preset health level, find the correction amount corresponding to the wear level according to the wear level and level parameter correction table.

[0103] Correction amounts refer to the values ​​used to adjust current process parameters, including additive and multiplicative correction amounts. The correction amount corresponding to the current wear level is obtained by consulting the level parameter correction table. Additive correction amounts are applicable to parameters adjusted using absolute values, such as back pressure, representing the increase or decrease in value based on the current value. Multiplicative correction amounts are applicable to parameters adjusted using proportional values, such as screw speed, representing the current value multiplied by the coefficient to obtain the corrected value. For example, for the wear level, the multiplicative correction amount for screw speed is 0.95, meaning the corrected speed is 95% of the current speed; the additive correction amount for back pressure is an increase of 2 bar, meaning the corrected back pressure is the current back pressure plus 2 bar.

[0104] Step 51: Obtain the current process parameters, and calculate the corrected key process parameters based on the correction amount.

[0105] Current process parameters refer to the set values ​​or actual operating values ​​of key process parameters under the current operating state of the injection molding machine. These values ​​are obtained by reading the set values ​​of each process parameter from the PLC controller. Screw speed is read from the speed register of the servo driver; back pressure is read from the set value register of the back pressure control unit; barrel temperature is read from the temperature set values ​​of each segment of the temperature control module; injection parameters are read from the parameter register of the injection control unit, including the injection speed curve, holding pressure, and holding time. All parameters are based on the current valid values ​​in the PLC controller.

[0106] Critical process parameters refer to the new process parameter values ​​obtained by calculating the current process parameters using correction amounts. For parameters using multiplicative correction amounts, the corrected value is equal to the current value multiplied by the multiplicative correction amount. Taking screw speed as an example, if the current speed is 80 revolutions per minute (RPM), note that the correction factor for the corresponding grade is 0.95, then the corrected speed is 76 RPM. For parameters using additive correction amounts, the corrected value is equal to the current value plus the additive correction amount. Taking back pressure as an example, if the current back pressure is 25 bar, note that the additive correction amount for the corresponding grade is an increase of 2 bar, then the corrected back pressure is 27 bar. For barrel temperature, the corrected value is equal to the current value plus the temperature correction amount; a negative temperature correction amount indicates a decrease in temperature.

[0107] Step 52: Send the key process parameters to the preset PLC controller to adjust the screw speed, back pressure, barrel temperature and injection parameters.

[0108] Following the instructions in step 7, the PLC controller sends the corrected key process parameters to the PLC controller via communication. The PLC controller then writes the parameters into the corresponding execution unit register, enabling the injection molding machine to operate according to the corrected parameters.

[0109] This also includes a safety constraint method for automatically adjusting key process parameters of the injection molding machine, which includes: Step 53: When the critical process parameter falls within the preset safety range, the current process parameter is gradually adjusted to the critical process parameter.

[0110] The safety range refers to the allowable numerical range for adjusting each process parameter. This range is determined by those skilled in the art based on the injection molding machine's specifications, screw material, and barrel pressure rating. Each process parameter has a separate upper and lower safety limit. For example: the lower limit for screw speed is 40 revolutions per minute, and the upper limit is 100 revolutions per minute; the lower limit for back pressure is 5 bar, and the upper limit is 40 bar; the single adjustment range for each section of the barrel temperature should not exceed 5 degrees Celsius, and the overall temperature fluctuation should not exceed 10 degrees Celsius; the lower limit for injection speed is 20 millimeters per second, and the upper limit is 80 millimeters per second.

[0111] Gradual adjustment refers to a method of gradually adjusting process parameters from their current values ​​to target values. First, the total adjustment amount from the current process parameter to the critical process parameter is calculated. This total adjustment amount is then broken down into multiple smaller steps, which are gradually completed over a certain number of work cycles. This number is preset by those skilled in the art. Each work cycle corresponds to a complete injection molding cycle. As an example, the number of gradual adjustment cycles can be set to 5 to 10 injection molding cycles. In each cycle, the system moves the current parameter closer to the target value by a step size equal to the total adjustment amount divided by the number of gradual adjustment cycles. Taking the screw speed adjustment from 80 rpm to 76 rpm with 5 gradual adjustment cycles as an example, the speed decreases by 0.8 rpm each cycle, reaching the target value of 76 rpm after 5 cycles. During the gradual adjustment process, the adjustment amount in each cycle is executed by the PLC controller.

[0112] When the critical process parameters fall within the preset safety range, it means that the process parameters can be adjusted. Therefore, the current process parameters are adjusted to the critical process parameters in the above gradual change method.

[0113] Step 54: Continuously obtain product quality indicators.

[0114] The product quality indicators refer to the quantitative parameters for evaluating whether an injection-molded product is qualified. The product quality indicators include injection volume deviation, consistency of the holding pressure curve, product weight deviation, and product size deviation, etc. The injection volume deviation is calculated by a screw position sensor, and the deviation value is obtained by comparing the actual injection stroke with the set stroke; the consistency of the holding pressure curve is collected by a cavity pressure sensor or a hydraulic pressure sensor and compared with a standard curve; the product weight deviation is obtained through an online weighing device or manual sampling measurement; the product size deviation is obtained through an online measuring device or off-line inspection. The above indicators are continuously collected, and the sampling frequency is synchronized with the injection molding cycle, and is updated once after each molding cycle ends.

[0115] Step 55: If the product quality indicators exceed the preset allowable range, the key process parameters are rolled back to the current process parameters before adjustment and a preset adjustment alarm signal is output.

[0116] The allowable range refers to the acceptable interval within which the product quality indicators are considered qualified, and is determined by those skilled in the art according to the quality standards and process requirements of the injection-molded product. An upper limit and a lower limit are set for each product quality indicator respectively.

[0117] The adjustment alarm signal refers to the warning signal when the product quality indicators exceed the allowable range, which is preset by those skilled in the art and sent to the human-machine interaction interface in the form of a pop-up window and audible and visual prompts to notify the operator. At the same time, the alarm information is recorded in the system log, including the alarm time, the quality indicators exceeding the range, the name of the parameter rolled back, and the parameter values before and after the rollback. The operator can view the alarm details through the human-machine interaction interface.

[0118] If the product quality indicators exceed the preset allowable range, it means that the product quality indicators are not within the acceptable interval of being qualified; therefore, the key process parameters are rolled back to the current process parameters before adjustment and a preset adjustment alarm signal is output.

[0119] If the product quality indicators do not exceed the preset allowable range, it means that the product quality indicators are within the acceptable interval of being qualified, and no additional operations are performed.

[0120] Among them, it also includes: Step 56: If the continuously predicted wear level is the preset alarm level, count the number of consecutive alarms with the wear level being the alarm level.

[0121] The alarm level refers to the highest level in the wear level, at this time the screw wear has reached a serious level, and there is a risk of failure if it continues to run. The alarm level is obtained through the output of the CNN-LSTM deep learning model.

[0122] The number of consecutive alarms refers to the cumulative number of times that the wear level is determined to be the alarm level during continuous prediction. By maintaining a counter, the number of predictions with the wear level continuously being the alarm level is recorded.

[0123] If the wear level predicted continuously is the preset alarm level, it indicates that the screw wear has been in a severe state and is not caused by occasional deviations; at this time, the system counts the number of times the alarm level occurs consecutively.

[0124] If the wear level predicted continuously does not match the preset alarm level, it means that the screw wear is not yet severe, and no additional operation is required.

[0125] Step 57: If the number of consecutive alarms exceeds the preset number of alarms and the remaining service life is lower than the preset shutdown threshold, trigger the preset shutdown protection signal.

[0126] The alarm count refers to the threshold number of consecutive occurrences of an alarm level that triggers the shutdown protection signal. The alarm count is determined by those skilled in the art based on the injection molding machine's safety redundancy and false alarm tolerance. For example, the alarm count can be set to 3 times.

[0127] The shutdown threshold refers to the critical value of remaining service life at which the shutdown protection signal is triggered. The shutdown threshold is determined by those skilled in the art based on the safety boundary of screw failure.

[0128] The shutdown protection signal is a control command signal that triggers an emergency shutdown of the injection molding machine. This signal is preset by those skilled in the art and is generated when the number of consecutive alarms exceeds a preset alarm count and the remaining service life is lower than a preset shutdown threshold. It is then sent to the PLC controller, which executes the emergency shutdown procedure, including stopping the screw rotation, stopping the injection action, and cutting off the power to the drive motor.

[0129] If the number of consecutive alarms exceeds the preset alarm count and the remaining service life is lower than the preset shutdown threshold, it indicates that the screw is close to failure and needs to be shut down immediately. Therefore, the shutdown protection signal is triggered, the human-machine interface displays the shutdown reason and alarm information, and the shutdown event is recorded to the system log.

[0130] If the number of consecutive alarms is not greater than the preset number of alarms or the remaining service life is not less than the preset shutdown threshold, it means that the conditions for forced shutdown have not yet been met. That is, the alarm level may be an occasional false alarm or the wear level has not yet endangered safe operation. At this time, continue to maintain the normal monitoring and adaptive control state, and continue to perform prediction and statistics at the original frequency.

[0131] The methods for outputting a preset maintenance prompt signal if the remaining service life is lower than a preset service life threshold include: Step 60: Calculate the latest maintenance time based on the remaining service life and the preset safety margin.

[0132] Safety margin refers to the proportion of the remaining useful life available for scheduled maintenance, and its value ranges from 0 to 1. Safety margin is preset by those skilled in the art.

[0133] The latest maintenance time refers to the latest maintenance time to ensure that maintenance is completed before the screw fails. The latest maintenance time is calculated based on the remaining service life and the preset safety margin. The latest maintenance time is equal to the current time plus the product of the remaining service life and the safety margin.

[0134] Step 61: Obtain the production schedule plan, find the maintainable time window in the production schedule plan based on the latest maintenance time, and generate a maintenance work order.

[0135] A production schedule is a production task arrangement table for the injection molding machine's production line, containing planned production tasks and downtime information for each time period. It is obtained from the Manufacturing Execution System (MES) by connecting to the MES. The production schedule includes planned production periods, planned downtime periods, and low-load production periods. For example, planned production periods can be normal production time, and planned downtime periods can be equipment maintenance, mold changeover time, or shift breaks. If not connected to the MES, operators can manually input or import production schedules through a human-machine interface.

[0136] The maintainable time window refers to the time period within the production schedule that can be used for screw maintenance operations. In the production schedule, select an idle time period no later than the latest maintenance time, prioritizing planned downtime periods such as equipment maintenance or mold change times. If no planned downtime periods are available, select low-load production periods or production gaps. If multiple time windows are available, the system recommends the one closest to the latest maintenance time.

[0137] A maintenance work order is a work document that schedules screw maintenance tasks. Maintenance work orders are automatically generated based on the remaining service life and production scheduling plan, recommending available maintenance time windows.

[0138] Step 62: Find the required quantity of spare parts based on the maintenance work order and check the current inventory.

[0139] Spare parts requirement refers to the number of spare parts needed to replace in one screw maintenance operation. The spare parts requirement is determined based on the maintenance operation type and screw specifications in the maintenance work order. When the maintenance operation is screw replacement, the spare parts requirement is one set, including one screw and its associated components; when the maintenance operation is partial maintenance, the spare parts requirement is determined based on the actual parts that need to be replaced.

[0140] Current inventory refers to the quantity of spare parts currently available in the warehouse. Spare part demand can be obtained from a pre-set spare parts configuration table, which is pre-established by those skilled in the art based on the injection molding machine model and screw specifications. The current inventory of spare parts can be queried by connecting to an enterprise resource planning (ERP) system. Query criteria include spare part name, specifications, and warehouse location. If not connected to an ERP system, operators can manually enter the current inventory quantity through a human-machine interface. Query results include available inventory quantity and in-transit inventory quantity; in-transit inventory quantity refers to the quantity of spare parts that have been purchased but not yet received into the warehouse.

[0141] Step 63: If the current inventory is less than the spare parts requirement, generate a spare parts purchase request and output a maintenance prompt signal.

[0142] A spare parts purchase requisition is a document requesting the purchase of spare screws. The required quantity of spare parts is determined based on the maintenance work order, and the current spare parts inventory is checked. If the inventory is less than the required quantity, a spare parts purchase requisition is automatically generated.

[0143] If the current inventory is less than the spare parts requirement, it means that the current inventory is insufficient for spare parts replacement. Therefore, a spare parts purchase request will be automatically generated and a maintenance reminder signal will be output. The purchase request includes: spare parts name, specifications, quantity to be purchased, and suggested delivery date. The quantity to be purchased is the spare parts requirement minus the current inventory. The suggested delivery date is no later than the start time of the recommended maintenance time window.

[0144] If the current inventory is not less than the spare parts demand, it means that the current inventory is sufficient to replace the spare parts, and no additional operation is required.

[0145] Among them, the method of recording the actual service life of the screw and feeding it back to the CNN-LSTM deep learning model for updating when a preset maintenance completion signal is received includes: Step 70: Define the final remaining useful life as the last predicted useful life before maintenance.

[0146] Step 71: Compare the actual service life with the final remaining service life and calculate the prediction deviation.

[0147] Prediction bias refers to the difference between the actual service life and the predicted total service life. The actual service life of the old screw recorded by the PLC controller is compared with the final remaining service life predicted by the CNN-LSTM deep learning model before maintenance, and the prediction bias is calculated. The prediction bias equals the actual service life minus the sum of the old screw's running time and the final remaining service life. A positive bias indicates that the model prediction is conservative, while a negative bias indicates that the prediction is aggressive. The calculation method for prediction bias is not limited to the above; it can also be achieved by directly comparing the actual service life with the initial predicted total service life, as long as the deviation in the model's prediction accuracy can be quantified.

[0148] Step 72: Feed the prediction bias back to the CNN-LSTM deep learning model to update the CNN-LSTM deep learning model.

[0149] The calculated prediction bias is used as feedback data and correlated with historical feature data collected during the operation of the old screw to form new training samples. New, accumulated data is uploaded periodically, such as quarterly, to incrementally train the CNN-LSTM deep learning model, adjusting it to achieve higher accuracy in subsequent predictions. After incremental training, the updated model is distributed to the edge computing unit to replace the original model.

[0150] Based on the same inventive concept, embodiments of the present invention provide a multi-dimensional fusion-based injection molding machine screw wear prediction and control system.

[0151] A multi-dimensional fusion-based injection molding machine screw wear prediction and control system includes: The acquisition module is used to acquire multi-dimensional sensor signals, material cleaning characteristics, current process parameters, product quality indicators, and production scheduling plans; The memory stores a computer program that can be loaded and executed by the processor, which is a method for predicting and controlling the wear of injection molding machine screws based on multi-dimensional fusion. The processor loads and executes programs from memory.

[0152] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional modules is used as an example. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. The specific working process of the system, device, and unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0153] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for predicting and controlling screw wear in injection molding machines based on multi-dimensional fusion, characterized in that, include: Step 1: Acquire multi-dimensional sensor signals during the operation of the injection molding machine screw and barrel; Step 2: Analyze and fuse the multidimensional sensor signals to generate the fused feature vector for the current moment; Step 3: Arrange the fused feature vectors of multiple consecutive moments within a preset time period in chronological order to construct the input time series; Step 4: Input the input time series into the preset CNN-LSTM deep learning model to predict the remaining service life and wear level; Step 5: Automatically adjust the key process parameters of the injection molding machine according to the wear level and the preset level parameter correction table; Step 6: If the remaining service life is lower than the preset service life threshold, output a preset maintenance prompt signal; Step 7: When the preset maintenance completion signal is received, record the actual service life of the screw and feed it back to the CNN-LSTM deep learning model for updating.

2. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 1, characterized in that, Methods for analyzing and fusing multidimensional sensor signals to generate a fused feature vector at the current moment include: Step 20: The multidimensional sensing signals include vibration signals, screw torque signals, barrel temperature signals, screw back pressure signals, and screw current signals; Step 21: Time synchronization of the multidimensional sensor signals to obtain time-aligned multi-channel signals; Step 22: Preprocess and extract features from the multi-channel signals to obtain multi-channel features, including vibration features, screw torque features, barrel temperature features, screw back pressure features, and screw current features; Step 23: Concatenate and reduce the dimensions of the multi-channel features to generate the fused feature vector at the current time.

3. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 2, characterized in that, Also includes: Step 24: When a preset cleaning start signal is received, a preset cleaning enhancement mode is used to acquire multi-dimensional sensing signals; Step 25: Obtain the unique cleaning characteristics during the cleaning process; Step 26: Combine the material cleaning features and multi-channel features to generate a fused feature vector.

4. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 3, characterized in that, It also includes a material cleaning optimization method when a preset material cleaning start signal is received, the method comprising: Step 240: Obtain the current wear level, the first melting temperature of the material before cleaning, and the second melting temperature of the material after cleaning; Step 241: Select the corresponding cleaning speed curve from the preset speed curve table according to the current wear level; Step 242: Calculate the temperature difference between the first melting temperature and the second melting temperature; Step 243: Calculate the segmented temperature transition curve based on the temperature difference and the preset temperature change rate; Step 244: Retrieve the cleaning time coefficient corresponding to the current wear level and multiply it by the preset basic cleaning time to calculate the suggested time value; Step 245: Send the material cleaning speed curve, segmented temperature transition curve, and suggested time value to the preset PLC controller and execute.

5. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 1, characterized in that, The method of automatically adjusting the key process parameters of the injection molding machine based on the wear level and the preset level parameter correction table also includes: Step 50: When the wear level is not the preset health level, find the correction amount corresponding to the wear level according to the wear level and level parameter correction table; Step 51: Obtain the current process parameters, and calculate the corrected key process parameters based on the correction amount. Step 52: Send the key process parameters to the preset PLC controller to adjust the screw speed, back pressure, barrel temperature and injection parameters.

6. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 5, characterized in that, It also includes a safety constraint method for automatically adjusting key process parameters of the injection molding machine, which includes: Step 53: When the critical process parameter falls within the preset safety range, the current process parameter is gradually adjusted to the critical process parameter. Step 54: Continuously obtain product quality indicators; Step 55: If the product quality index exceeds the preset allowable range, revert the key process parameters to the current process parameters before adjustment and output the preset adjustment alarm signal.

7. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 6, characterized in that, Also includes: Step 56: If the wear level obtained by continuous prediction is the preset alarm level, count the number of consecutive alarms when the wear level is the alarm level. Step 57: If the number of consecutive alarms exceeds the preset number of alarms and the remaining service life is lower than the preset shutdown threshold, trigger the preset shutdown protection signal.

8. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 1, characterized in that, If the remaining service life is lower than the preset service life threshold, the methods for outputting a preset maintenance prompt signal include: Step 60: Calculate the latest maintenance time based on the remaining service life and the preset safety margin; Step 61: Obtain the production schedule plan, find the maintainable time window in the production schedule plan based on the latest maintenance time, and generate a maintenance work order; Step 62: Find the required quantity of spare parts based on the maintenance work order and check the current inventory. Step 63: If the current inventory is less than the spare parts requirement, generate a spare parts purchase request and output a maintenance prompt signal.

9. The method for predicting and controlling screw wear of injection molding machines based on multi-dimensional fusion according to claim 8, characterized in that, The methods for recording the actual service life of the screw and feeding it back to the CNN-LSTM deep learning model for updating when a preset maintenance completion signal is received include: Step 70: Define the final remaining useful life as the last predicted useful life before maintenance; Step 71: Compare the actual service life with the final remaining service life and calculate the prediction deviation; Step 72: Feed the prediction bias back to the CNN-LSTM deep learning model to update the CNN-LSTM deep learning model.

10. A multi-dimensional fusion-based injection molding machine screw wear prediction and control system, characterized in that, include: The acquisition module is used to acquire multi-dimensional sensor signals, material cleaning characteristics, current process parameters, product quality indicators, and production scheduling plans; A memory for storing the program of the injection molding machine screw wear prediction and control method based on multi-dimensional fusion as described in any one of claims 1 to 9; The processor loads and executes programs from memory.

Images