Remote control system and method for injection molding equipment

Abstract

The present invention relates to the technical field of injection molding equipment, and in particular to a remote control system and method for injection molding equipment, which collects the operating parameters, finished product characterization images and weight of the injection molding equipment in real time, covering key operating parameters such as injection speed, holding pressure parameters, and clamping force; obtains geographic location data, and divides the equipment into location zones through the K-means clustering algorithm, taking into account the correlation of the injection molding equipment and using convolutional neural networks to determine the consistency of finished products produced by equipment in the same area, ensuring stable product quality, and improving management accuracy through correlation analysis of equipment operating parameters. The remote control module uses a feedforward neural network to analyze operating parameter deviations, combines historical fault data, and generates a total fault score through weighted calculation to achieve intelligent judgment of equipment failures. It improves the timeliness of fault warnings, avoids the lag of manual monitoring, and can automatically control the equipment operating status according to the degree of fault impact, thereby improving the stability and efficiency of the production line.

Description

Technical Field

[0001] The present invention relates to the technical field of injection molding equipment, and in particular to a remote control system and method for injection molding equipment. Background Art

[0002] Injection molding equipment, also known as injection molding machine or injection machine, is the main molding equipment that uses plastic molding molds to make plastic products of various shapes from thermoplastics or thermosetting plastics. It is composed of core components such as hydraulic transmission mechanism, injection mechanism, electronic control mechanism, and clamping mechanism. These components usually reflect operating parameters including injection speed, holding pressure parameters, clamping force, sol time, sol load pressure, injection peak pressure, and injection end point. Slight changes in these parameters may affect the quality of the final product. At the same time, various component failures may occur in the long-term use of injection molding equipment. If these failures are not discovered and handled in time, they may have a serious impact on the efficiency of the production line and product quality.

[0003] Traditional injection molding equipment management and control methods mainly rely on manual inspections and regular maintenance, which is inefficient and cannot monitor equipment status in real time, making it difficult to detect and respond to potential equipment failures in a timely manner. In addition, as the scale of production expands, companies often have a large number of injection molding equipment, which are distributed in different production areas, making equipment management and maintenance more complicated. Summary of the invention

[0004] To solve the above problems, the present invention provides a remote control system and method for injection molding equipment. By associating the injection molding equipment and performing artificial intelligence processing on the associated equipment, the timeliness of fault warning is improved, the lag in manual monitoring is avoided, and the operating status of the injection molding equipment is automatically controlled according to the production impact of the fault, thereby improving the stability and efficiency of the production line.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] A remote control system for injection molding equipment, comprising: a data acquisition module, an injection molding equipment positioning module, an injection molding equipment association module and a remote control module, wherein the data acquisition module is connected to the injection molding equipment association module and the remote control module respectively, and the injection molding equipment positioning module, the injection molding equipment association module and the remote control module are connected in sequence;

[0007] The data acquisition module is used to acquire the operating parameters of the injection molding equipment, the characterization image of the injection molded product and the weight of the injection molded product, wherein the operating parameters include injection speed, holding pressure parameter, clamping force, sol time, sol load pressure, injection peak pressure and injection end point;

[0008] The injection molding equipment positioning module is used to obtain the positioning data of the injection molding equipment;

[0009] The injection molding equipment association module is used to classify the injection molding equipment into position zones according to the positioning data of the injection molding equipment, determine whether the injection molded products output by the injection molding equipment in the same position zone are the same through a convolutional neural network, and associate the injection molding equipment according to the determination result, the weight of the injection molded products and the operating parameters;

[0010] The remote control module is used to perform the following steps:

[0011] S1. Generate an operating parameter deviation vector according to the difference between the operating parameter of the injection molding equipment to be determined and the preset threshold value of the operating parameter;

[0012] S2, outputting the real-time score of each fault type through a feedforward neural network based on the operating parameter deviation vector;

[0013] S3, calculating the frequency ratio of the historical fault types of the associated injection molding equipment of the injection molding equipment to be determined, and standardizing and generating the historical weight of each fault type;

[0014] S4. Weighting the real-time score of each fault type and the historical weight of each fault type to generate a total score of each fault type, and determining the equipment fault according to the total score of each fault type;

[0015] S5. Match the production impact degree according to the equipment failure, remotely control the operation status of the injection molding equipment according to the production impact degree and issue a fault alarm prompt.

[0016] Furthermore, the step of classifying the injection molding equipment into position zones according to the positioning data of the injection molding equipment comprises the following steps:

[0017] Acquire positioning data of the injection molding equipment, wherein the positioning data includes latitude and longitude coordinate information of the injection molding equipment;

[0018] The injection molding equipment is clustered using K-means based on its longitude and latitude coordinate information and classified into several locations.

[0019] Furthermore, the training steps of the convolutional neural network include:

[0020] Obtaining representation images of injection molded products output by injection molding equipment in different position areas for normalization and image enhancement;

[0021] Label the data of similar injection molded products and build a training set;

[0022] Construct a convolutional neural network model, take the training set as input, and train the output to determine whether the injection molded products are the same;

[0023] Through the cross entropy loss function and stochastic gradient descent, back propagation adjusts the network weight parameters, and iterative training is carried out until the model converges.

[0024] Furthermore, associating the injection molding equipment according to the determination result, the weight of the injection molded product and the operating parameters includes:

[0025] For two injection molding devices with the same injection molded product as the result, the two injection molding devices are associated by adding the absolute value of the difference in weight of the injection molded product and the absolute value of the difference in operating parameters.

[0026] Further, the S2 comprises the following steps:

[0027] The normalized operating parameter deviation vector is used as input and input into the input layer of the feedforward neural network;

[0028] Through the hidden layer of the feedforward neural network, the input operating parameter deviation vector is nonlinearly mapped to generate a preliminary fault type score;

[0029] The output layer of the feedforward neural network is processed by activation function to generate real-time scores for each fault type.

[0030] Furthermore, the formula of the feedforward neural network is as follows:

[0031] ;

[0032] in, is the real-time score of the i-th fault type; is the jth operating parameter deviation input; For connection input and the weight of the i-th output node; is the bias associated with the i-th output node; is the activation function.

[0033] Furthermore, the loss function for training the feedforward neural network is as follows:

[0034] ;

[0035] in, is the loss function; is the sample size; is the category number of the fault type; For sample

[0036] The true label of i in category k; is the predicted output of the feedforward neural network during training; is the regularization coefficient; is the weight matrix of the lth layer.

[0037] Further, determining the equipment failure according to the total score of each fault type includes:

[0038] The total score of each fault type is compared with the preset fault threshold. If the total score of only one fault type exceeds its corresponding threshold, it is determined that the device has this fault type. If the total score of more than one fault type exceeds its threshold at the same time, the fault types are sorted according to their total scores, and it is determined that the device has the fault type with the highest total score.

[0039] A remote control method for injection molding equipment, applied to any of the remote control systems for injection molding equipment described above, comprises the following steps:

[0040] Obtaining the operating parameters of the injection molding equipment, the characterization image of the injection molding product and the weight of the injection molding product, wherein the operating parameters include injection speed, holding pressure parameter, clamping force, sol time, sol load pressure, injection peak pressure and injection end point; obtaining the positioning data of the injection molding equipment;

[0041] The injection molding equipment is divided into position areas according to the positioning data of the injection molding equipment, and the injection molding products output by the injection molding equipment in the same position area are judged to be the same through the convolutional neural network, and the injection molding equipment is associated according to the judgment result, the weight of the injection molding products and the operating parameters;

[0042] Generate an operating parameter deviation vector according to the difference between the operating parameter of the injection molding equipment to be determined and a preset threshold value of the operating parameter;

[0043] Based on the operating parameter deviation vector, the real-time score of each fault type is output through a feed-forward neural network;

[0044] Calculate the frequency ratio of the historical fault types of the associated injection molding equipment to be determined, and standardize and generate the historical weight of each fault type;

[0045] The real-time score of each fault type and the historical weight of each fault type are weighted to generate a total score of each fault type, and the equipment fault is determined based on the total score of each fault type;

[0046] According to the degree of production impact matched by equipment failure, the operating status of the injection molding equipment is remotely controlled and a fault alarm is issued according to the degree of production impact.

[0047] The beneficial effect of the present invention is that the data acquisition module of the present invention collects the operating parameters of the injection molding equipment, the characterization image of the injection molding product and the weight of the finished product in real time, including key operating parameters including injection speed, holding pressure parameters, clamping force, sol time, sol load pressure, injection peak pressure and injection end point. The injection molding equipment positioning module obtains the geographical location data of the equipment, and classifies the equipment based on these positioning data through the injection molding equipment association module, and uses the K-means clustering algorithm to effectively divide and manage the equipment distributed in different areas, solving the management problem caused by the complex distribution of equipment in large-scale production. At the same time, the convolutional neural network is used to determine the consistency of the finished products produced by the equipment in the same area, ensure the stability of the quality of the finished products, and further improve the accuracy of equipment management through the correlation analysis of the operating parameters. The remote control module analyzes the deviation between the operating parameters of the associated injection molding equipment and the preset threshold through the feedforward neural network, outputs the real-time score of each fault type, and combines the historical fault data of the associated injection molding equipment to generate the total score of each fault type through weighted analysis, and finally realizes the intelligent determination of equipment failure. This intelligent diagnosis method greatly improves the timeliness of fault warning, avoids the lag in manual monitoring, and can automatically control the operating status of the injection molding equipment according to the degree of production impact of the fault, thereby improving the stability and efficiency of the production line. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a structural schematic diagram of a remote control system for injection molding equipment in the present invention.

[0049] Figure 2 It is a flow chart of the execution steps of the remote control module in the present invention. DETAILED DESCRIPTION

[0050] See also Figure 1-2 As shown, the present invention relates to a remote control system for injection molding equipment, comprising: a data acquisition module, an injection molding equipment positioning module, an injection molding equipment association module and a remote control module, wherein the data acquisition module is connected to the injection molding equipment association module and the remote control module respectively, and the injection molding equipment positioning module, the injection molding equipment association module and the remote control module are connected in sequence;

[0051] The data acquisition module is used to acquire the operating parameters of the injection molding equipment, the characterization image of the injection molded product and the weight of the injection molded product, wherein the operating parameters include injection speed, holding pressure parameter, clamping force, sol time, sol load pressure, injection peak pressure and injection end point;

[0052] The injection molding equipment positioning module is used to obtain the positioning data of the injection molding equipment;

[0053] The injection molding equipment association module is used to classify the injection molding equipment into position zones according to the positioning data of the injection molding equipment, determine whether the injection molded products output by the injection molding equipment in the same position zone are the same through a convolutional neural network, and associate the injection molding equipment according to the determination result, the weight of the injection molded products and the operating parameters;

[0054] The remote control module is used to perform the following steps:

[0055] S1. Generate an operating parameter deviation vector according to the difference between the operating parameter of the injection molding equipment to be determined and the preset threshold value of the operating parameter;

[0056] S2, outputting the real-time score of each fault type through a feedforward neural network based on the operating parameter deviation vector;

[0057] S3, calculating the frequency ratio of the historical fault types of the associated injection molding equipment of the injection molding equipment to be determined, and standardizing and generating the historical weight of each fault type;

[0058] S4. Weighting the real-time score of each fault type and the historical weight of each fault type to generate a total score of each fault type, and determining the equipment fault according to the total score of each fault type;

[0059] S5. Match the production impact degree according to the equipment failure, remotely control the operation status of the injection molding equipment according to the production impact degree and issue a fault alarm prompt.

[0060] It should be noted that the data acquisition module realizes real-time monitoring of the operating status of the injection molding equipment through sensors and controllers installed on the injection molding equipment. Specifically, the data acquisition module includes hardware devices such as high-speed data acquisition cards, temperature sensors, and pressure sensors, which are used to obtain the injection speed, holding pressure parameters, clamping force, sol time, sol load pressure, injection peak pressure, and injection end point of the injection molding equipment. At the same time, the image acquisition unit uses a high-definition camera or an industrial camera to capture the characterization image of the injection molded product in real time for subsequent quality evaluation. The injection molding equipment positioning module obtains the geographical location information of the injection molding equipment through a GPS module or ultra-wideband (UWB) positioning technology to provide accurate equipment positioning services. The hardware architecture of this module includes a high-precision positioning receiver and antenna, which continuously updates the positioning information of the equipment through data transmission with the control system inside the equipment. The injection molding equipment association module uses a processor and a dedicated computing unit to execute the K-means clustering algorithm based on the acquired positioning data to classify the injection molding equipment distributed in different areas. The module also contains a dedicated neural network acceleration chip for inference calculation of convolutional neural networks to determine whether the injection molding products produced by different equipment in the same area are consistent. Through this judgment, different devices are associated through the equipment operating parameters and the injection molding product quality system to optimize the production management of the equipment. The remote control module executes the core control logic of the system through industrial control computers, servers or edge computing devices. First, the processor compares the collected operating parameters with the preset thresholds to generate an operating parameter deviation vector. The deviation vector is input to the feedforward neural network accelerator in the remote control module, and the accelerator completes the real-time analysis of the deviation vector and outputs the real-time score of each fault type. Then, the system further calculates the historical fault frequency ratio of the injection molding equipment to be determined, and generates the historical weight of each fault type through standardization. The purpose of standardization is to eliminate the magnitude difference in the frequency of occurrence of different fault types so that these data can be compared on the same scale. Standardization usually includes subtracting the mean from the frequency ratio and dividing it by the standard deviation. The remote control module also performs a weighted operation on the real-time score and the historical weight through a control algorithm to generate a total score for each fault type. If the total score exceeds the preset threshold, the system determines that the equipment is faulty. After the fault is determined, the system automatically adjusts the operating status of the injection molding equipment according to the impact of the fault on production, issues a fault alarm if necessary, and triggers an emergency plan.

[0061] Furthermore, the step of classifying the injection molding equipment into position zones according to the positioning data of the injection molding equipment comprises the following steps:

[0062] Acquire positioning data of the injection molding equipment, wherein the positioning data includes latitude and longitude coordinate information of the injection molding equipment;

[0063] The injection molding equipment is clustered using K-means based on its longitude and latitude coordinate information and classified into several locations.

[0064] In some embodiments, the geographic location information of each injection molding device, including specific longitude and latitude coordinate data, is first obtained from the positioning module. These data represent the distribution of all devices in the production area. The system extracts longitude and latitude coordinates from the positioning data of the injection molding device and constructs a set of data points in a two-dimensional space. Each data point represents the location of an injection molding device. In order to preliminarily divide the location area, the system needs to set the number of clusters K, which is usually pre-set according to the size of the production area and the equipment density, or can be dynamically determined by a heuristic method. In the initialization phase, the K-means algorithm randomly selects K initial cluster centers. Each cluster center represents a potential equipment cluster area. Next, the system calculates the Euclidean distance between each injection molding device and these initial cluster centers. The system assigns each device to the area where the nearest cluster center is located based on the minimum distance. After completing the initial assignment, the system updates the center point of each cluster. This is achieved by calculating the average of the longitude and latitude coordinates of all devices in each area. This process continues in an iterative manner. In each iteration, the system recalculates the distance between the device and the updated cluster center and assigns the device again. After each allocation is completed, the coordinates of the cluster center will be updated until one of the following stop conditions is met: 1. The moving distance of the cluster center is less than the preset convergence threshold. 2. The maximum number of iterations is reached. The core goal of the K-means algorithm is to minimize the sum of the distances from the internal devices to the center of each cluster. This method allocates devices to the most suitable location area through iterative optimization. During the clustering process, in order to improve computing efficiency, the system usually uses parallel computing resources, such as GPU or multi-threaded CPU, to achieve rapid clustering of a large number of devices. Through this method, the system can automatically identify natural equipment clusters within the production area and divide these devices into different location areas.

[0065] Furthermore, the training steps of the convolutional neural network include:

[0066] Obtaining representation images of injection molded products output by injection molding equipment in different position areas for normalization and image enhancement;

[0067] Label the data of similar injection molded products and build a training set;

[0068] Construct a convolutional neural network model, take the training set as input, and train the output to determine whether the injection molded products are the same;

[0069] Through the cross entropy loss function and stochastic gradient descent, back propagation adjusts the network weight parameters, and iterative training is carried out until the model converges.

[0070] It should be noted that the acquired image data usually has different resolutions and lighting conditions. Therefore, data preprocessing must be performed before training. The first operation is image normalization, which adjusts the pixel values ​​of each image to a uniform range, usually between 0 and 1. Normalization helps to eliminate the effects of lighting and contrast under different shooting conditions, so that the image data is processed at the same scale, which is conducive to improving the training effect of the model. Next, the system will perform image enhancement operations, and artificially generate additional training samples by rotating, flipping, scaling, translating, and other operations on the image. These enhanced images help the model improve its adaptability to sample diversity during training and reduce the risk of overfitting. After completing image preprocessing, the system enters the data annotation stage. Data annotation is manually completed by experienced quality inspection engineers to ensure that each image is accurately labeled. The label content usually indicates the type of the finished product or whether it is the same. These labeled data sets will be divided into training sets and validation sets by the system. The training set is used for model learning, and the validation set is used for model performance evaluation. Next, the system builds a convolutional neural network (CNN) model. Convolutional neural networks consist of multiple layers of convolutional layers, pooling layers, and fully connected layers. In the convolutional layer, the system uses multiple convolution kernels to extract local features from the image. These convolution kernels are trained to capture important features such as edges and textures in the image. By using convolution kernels of different sizes, the network can understand the details of the image at different scales. The convolutional feature map enters the pooling layer, which further reduces the size of the feature map through downsampling operations (such as maximum pooling or average pooling), reduces the computational complexity, and retains the main features. This process can effectively reduce network parameters and avoid overfitting problems. After feature extraction, the feature map is flattened into a one-dimensional vector and input into the fully connected layer. The fully connected layer maps the high-dimensional feature vector to a fixed-length output vector, which represents the probability distribution of each category. The output layer usually uses a softmax function to convert the model prediction results into probabilities, which are used to determine whether they are finished products of the same type. During the training process, the system uses the cross entropy loss function as the objective function to evaluate the difference between the model output and the actual label. The cross entropy loss function has a high performance in classification tasks. It guides the optimization of model parameters by quantifying the credibility of model predictions. Specifically, the smaller the loss function value, the higher the accuracy of model predictions. To optimize the parameters of the model, the system uses the stochastic gradient descent (SGD) algorithm. SGD gradually updates the model weights by calculating the gradient of small batches of samples. In each iteration, the system calculates the gradient of the loss function relative to each weight parameter through the back-propagation algorithm, and adjusts the weights based on this gradient information, so that the prediction error of the model gradually decreases. The back-propagation process is the core step of training CNN. It gradually transfers the error from the output layer to the input layer through the chain rule to update the weights of all layers.During the training process, the system will adjust hyperparameters such as learning rate and batch size according to the performance of the validation set to ensure that the model converges quickly and reaches the optimal solution during the training process. To prevent the model from overfitting, the system also applies regularization techniques (such as L2 regularization) and dropout layers to randomly ignore the output of some neurons during training to improve the generalization ability of the model.

[0071] Furthermore, associating the injection molding equipment according to the determination result, the weight of the injection molded product and the operating parameters includes:

[0072] For two injection molding devices with the same injection molded product as the result, the two injection molding devices are associated by adding the absolute value of the difference in weight of the injection molded product and the absolute value of the difference in operating parameters.

[0073] It should be noted that the system obtains the consistency judgment results of the convolutional neural network model for the injection molded products. For two injection molding devices that produce the same judgment results in the same position area, their finished products have been judged by the convolutional neural network to have similar appearance and quality characteristics. However, visual judgment alone is not enough for a comprehensive evaluation, so it is necessary to further perform correlation analysis through the weight of the finished products and the operating parameters. To this end, the system extracts the weight data of the injection molded products corresponding to the two injection molding devices from the data acquisition module. The weight of the injection molded products is recorded in real time by the weighing module, and the system inputs these data into the analysis module to calculate the absolute value of the weight difference of the finished products produced by the two devices. The weight difference can reflect the problems such as material supply fluctuations or unstable process parameters that may exist in the production process, so this difference is used as one of the important indicators of equipment association. At the same time, the system also extracts key parameters closely related to the injection molding process from the operating parameter data of the equipment. These parameters are usually collected in real time by sensors on the equipment and stored in the system's database. The system calculates the difference of these operating parameters one by one and extracts the absolute value of the difference of each operating parameter between the two devices. Through this analysis step, the system can identify the difference in operating parameters of the equipment. Next, the system summarizes the absolute values ​​of the weight differences of the injection molded finished products and the absolute values ​​of the operating parameter differences. Specifically, the system weights and sums all the calculated absolute values ​​of the differences to generate a comprehensive difference index. This comprehensive difference index not only reflects the difference in the weight of the finished products, but also takes into account the fluctuations of various key operating parameters of the equipment during the production process. Therefore, this comprehensive difference index can fully reflect the correlation between the equipment. The system conducts a correlation evaluation on the two injection molding equipment based on the comprehensive difference index. When the comprehensive difference index is small, it means that the two equipment are relatively close in production performance and output quality. The system will mark them as highly correlated equipment. In subsequent equipment scheduling, data analysis, resource allocation and maintenance arrangements, the system will give priority to treating these equipment as a group to ensure production consistency and stability.

[0074] Further, the S2 comprises the following steps:

[0075] The normalized operating parameter deviation vector is used as input and input into the input layer of the feedforward neural network;

[0076] Through the hidden layer of the feedforward neural network, the input operating parameter deviation vector is nonlinearly mapped to generate a preliminary fault type score;

[0077] The output layer of the feedforward neural network is processed by activation function to generate real-time scores for each fault type.

[0078] In some embodiments, first, the system collects the operating parameters of the injection molding equipment, which include the injection speed, holding pressure parameters, clamping force, sol time, sol load pressure, injection peak pressure and injection end point of the injection molding equipment. The collected operating parameters will be normalized to convert all parameters into the same numerical range, usually with a normalization range of 0 to 1. This step is intended to eliminate the influence of different parameter dimensions, ensure the uniformity of input data, and improve the processing efficiency and accuracy of the feedforward neural network. The normalized operating parameter deviation vector will be used as the input of the feedforward neural network. The vector represents the difference between the operating state of the current injection molding equipment to be determined and the normal operating state. Specifically, the system inputs the normalized operating parameter deviation vector into the input layer of the feedforward neural network. The input layer of the feedforward neural network consists of multiple nodes, each of which corresponds to a deviation value of an operating parameter. This layer is responsible for receiving all normalized input data and passing it to subsequent network layers for processing. Next, the data of the input layer is nonlinearly mapped through several hidden layers. The hidden layer is usually composed of multiple neurons (nodes), and each neuron is connected to the nodes of the previous and next layers through a fully connected method with an activation function. In the hidden layer, the system converts the input data into high-order features through weighted calculations and activation function processing. The output of each neuron is a nonlinear mapping of the weighted sum of its inputs. Common activation functions include ReLU (Rectified Linear Unit) or Sigmoid function. Through the processing of the hidden layer, the neural network can extract complex patterns and features in the deviation of operating parameters, which are closely related to the potential fault type of the equipment. As the data is passed layer by layer in the feedforward neural network, the output of the hidden layer is gradually aggregated into a preliminary fault type score. Each fault type score represents the probability of a specific fault occurring in the equipment. After further processing, these scores will be used to evaluate the operating status of the equipment. In the last layer of the network, the output layer, the system applies the activation function to the output of the hidden layer to generate the final real-time score of the fault type. The choice of activation function usually depends on the specific application scenario. For example, the softmax function is often used for multi-classification problems. It can convert the output into a probability distribution to represent the relative probability of each fault type. The system uses these real-time scores to determine the current status of the device and provide a basis for subsequent decisions.

[0079] Furthermore, the formula of the feedforward neural network is as follows:

[0080] ;

[0081] in, is the real-time score of the i-th fault type; is the jth operating parameter deviation input; For connection input and the weight of the i-th output node; is the bias associated with the i-th output node; is the activation function.

[0082] Specifically, during this calculation, the input operating parameter deviation Will pass with weight The weighted sum is then added with the bias , constitutes the preliminary score of the i-th fault type, and then the activation function This linear combination is mapped into a nonlinear space so that the feedforward neural network can better capture complex fault modes. Common activation functions include ReLU (Rectified Linear Unit) or Sigmoid function, and the choice depends on the application requirements of fault judgment.

[0083] Furthermore, the loss function for training the feedforward neural network is as follows:

[0084] ;

[0085] in, is the loss function; is the sample size; is the category number of the fault type; is the true label of sample i in category k; is the predicted output of the feedforward neural network during training; is the regularization coefficient; is the weight matrix of the lth layer.

[0086] Specifically, the first part of the loss function is the cross entropy loss, which is used to measure the difference between the model's predicted output and the true label. Cross entropy loss is widely used in classification problems and can effectively quantify the degree of match between the model output and the target label. The second part of the loss function is the L2 regularization term, which imposes a penalty on the model's weight matrix to limit the weights from being too large and prevent the model from overfitting to specific features during training. The regularization coefficient λ controls the strength of the regularization term and is usually adjusted to a suitable value through experiments. By minimizing this loss function, the system can gradually optimize the weight parameters of the feedforward neural network during the training process, ultimately improving the prediction accuracy and generalization ability of the model in the fault type classification task.

[0087] Further, determining the equipment failure according to the total score of each fault type includes:

[0088] The total score of each fault type is compared with the preset fault threshold. If the total score of only one fault type exceeds its corresponding threshold, it is determined that the device has this fault type. If the total score of more than one fault type exceeds its threshold at the same time, the fault types are sorted according to their total scores, and it is determined that the device has the fault type with the highest total score.

[0089] In some embodiments, first, the system analyzes the operating parameters of the injection molding equipment through a feedforward neural network to generate real-time scores for each fault type. These scores represent the possibility of a specific fault occurring in the equipment. Subsequently, the system weights these real-time scores based on historical data and fault frequency to generate the final total score for each fault type. The total score for each fault type represents the comprehensive risk of the equipment on that fault type. In the step of determining the equipment fault, the system first compares the total score of these fault types with a preset fault threshold. The preset fault threshold is manually set based on the historical operation data of the equipment and industry standards, and is used to define the normal and abnormal states of the equipment under a specific fault type. If only the total score of one fault type exceeds the corresponding fault threshold, the system determines that the equipment has a fault of that type. This determination of a single fault type can quickly locate the source of the problem, and the system will trigger the corresponding fault handling process based on this judgment, such as suspending the operation of the equipment or issuing a fault alarm. However, in some cases, the total scores of multiple fault types may exceed their thresholds at the same time. At this time, the system needs to further process these fault types that exceed the threshold. To this end, the system will sort the fault types according to their total scores, and prioritize the fault types with the highest total scores as the primary fault types. The higher the total score, the greater the impact of the fault type on the equipment's operating status and the higher the probability of its occurrence. Once the primary fault type is determined, the system will mark other fault types that exceed the threshold but have lower scores as secondary fault types. This classification can help operation and maintenance personnel pay attention to secondary faults while handling primary faults, ensuring full recovery of the equipment.

[0090] The present invention also includes a remote control method for injection molding equipment, comprising the following steps:

[0091] Obtaining the operating parameters of the injection molding equipment, the characterization image of the injection molding product and the weight of the injection molding product, wherein the operating parameters include injection speed, holding pressure parameter, clamping force, sol time, sol load pressure, injection peak pressure and injection end point; obtaining the positioning data of the injection molding equipment;

[0092] The injection molding equipment is divided into position areas according to the positioning data of the injection molding equipment, and the injection molding products output by the injection molding equipment in the same position area are judged to be the same through the convolutional neural network, and the injection molding equipment is associated according to the judgment result, the weight of the injection molding products and the operating parameters;

[0093] Generate an operating parameter deviation vector according to the difference between the operating parameter of the injection molding equipment to be determined and a preset threshold value of the operating parameter;

[0094] Based on the operating parameter deviation vector, the real-time score of each fault type is output through a feed-forward neural network;

[0095] Calculate the frequency ratio of the historical fault types of the associated injection molding equipment to be determined, and standardize and generate the historical weight of each fault type;

[0096] The real-time score of each fault type and the historical weight of each fault type are weighted to generate a total score of each fault type, and the equipment fault is determined based on the total score of each fault type;

[0097] According to the degree of production impact matched by equipment failure, the operating status of the injection molding equipment is remotely controlled and a fault alarm is issued according to the degree of production impact.

[0098] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A remote control system for injection molding equipment, characterized in that: include: A data acquisition module, an injection molding equipment positioning module, an injection molding equipment association module and a remote control module, wherein the data acquisition module is connected to the injection molding equipment association module and the remote control module respectively, and the injection molding equipment positioning module, the injection molding equipment association module and the remote control module are connected in sequence; The data acquisition module is used to acquire the operating parameters of the injection molding equipment, the characterization image of the injection molded product and the weight of the injection molded product, wherein the operating parameters include injection speed, holding pressure parameter, clamping force, sol time, sol load pressure, injection peak pressure and injection end point; The injection molding equipment positioning module is used to obtain the positioning data of the injection molding equipment; The injection molding equipment association module is used to classify the injection molding equipment into position zones according to the positioning data of the injection molding equipment, determine whether the injection molded products output by the injection molding equipment in the same position zone are the same through a convolutional neural network, and associate the injection molding equipment according to the determination result, the weight of the injection molded products and the operating parameters; The remote control module is used to perform the following steps: S1. Generate an operating parameter deviation vector according to the difference between the operating parameter of the injection molding equipment to be determined and the preset threshold value of the operating parameter; S2, outputting the real-time score of each fault type through a feedforward neural network based on the operating parameter deviation vector; S3, calculating the frequency ratio of the historical fault types of the associated injection molding equipment of the injection molding equipment to be determined, and standardizing and generating the historical weight of each fault type; S4. Weighting the real-time score of each fault type and the historical weight of each fault type to generate a total score of each fault type, and determining the equipment fault according to the total score of each fault type; S5. Match the production impact degree according to the equipment failure, remotely control the operation status of the injection molding equipment according to the production impact degree and issue a fault alarm prompt; The method of classifying the location area of ​​the injection molding equipment according to the positioning data of the injection molding equipment comprises the following steps: Acquire positioning data of the injection molding equipment, wherein the positioning data includes latitude and longitude coordinate information of the injection molding equipment; Based on the longitude and latitude coordinate information of the injection molding equipment, the injection molding equipment is clustered using K-means and classified into several locations; The training steps of the convolutional neural network include: Obtaining representation images of injection molded products output by injection molding equipment in different position areas for normalization and image enhancement; Label the data of similar injection molded products and build a training set; Construct a convolutional neural network model, take the training set as input, and train the output to determine whether the injection molded products are the same; Through the cross entropy loss function and stochastic gradient descent, back propagation adjusts the network weight parameters, and iterates the training until the model converges; The method of associating the injection molding equipment according to the determination result, the weight of the injection molded product and the operating parameters includes: For two injection molding devices with the same injection molding product as determined, the two injection molding devices are associated by adding the absolute value of the difference in weight of the injection molding product and the absolute value of the difference in operating parameters; Determining equipment failure according to the total score of each fault type includes: The total score of each fault type is compared with the preset fault threshold. If the total score of only one fault type exceeds its corresponding threshold, it is determined that the device has this fault type. If the total score of more than one fault type exceeds its threshold at the same time, the fault types are sorted according to their total scores, and it is determined that the device has the fault type with the highest total score.

2. A remote control system for injection molding equipment according to claim 1, characterized in that: The S2 comprises the following steps: The normalized operating parameter deviation vector is used as input and input into the input layer of the feedforward neural network; Through the hidden layer of the feedforward neural network, the input operating parameter deviation vector is nonlinearly mapped to generate a preliminary fault type score; The output layer of the feedforward neural network is processed by activation function to generate real-time scores for each fault type.

3. A remote control system for injection molding equipment according to claim 2, characterized in that: The formula of the feedforward neural network is as follows: ； in, is the real-time score of the i-th fault type; is the jth operating parameter deviation input; For connection input and the weight of the i-th output node; is the bias associated with the i-th output node; is the activation function.

4. A remote control system for injection molding equipment according to claim 2, characterized in that: The loss function for training the feedforward neural network is as follows: ； in, is the loss function; is the sample size; is the category number of the fault type; is the true label of sample i in category k; is the predicted output of the feedforward neural network during training; is the regularization coefficient; is the weight matrix of the lth layer.

5. A remote control method for injection molding equipment, applied to a remote control system for injection molding equipment according to any one of claims 1 to 4, characterized in that: The following steps are involved: Obtaining the operating parameters of the injection molding equipment, the characterization image of the injection molding product and the weight of the injection molding product, wherein the operating parameters include injection speed, holding pressure parameter, clamping force, sol time, sol load pressure, injection peak pressure and injection end point; obtaining the positioning data of the injection molding equipment; The injection molding equipment is divided into position areas according to the positioning data of the injection molding equipment, and the injection molding products output by the injection molding equipment in the same position area are judged to be the same through the convolutional neural network, and the injection molding equipment is associated according to the judgment result, the weight of the injection molding products and the operating parameters; Generate an operating parameter deviation vector according to the difference between the operating parameter of the injection molding equipment to be determined and a preset threshold value of the operating parameter; Based on the operating parameter deviation vector, the real-time score of each fault type is output through a feed-forward neural network; Calculate the frequency ratio of the historical fault types of the associated injection molding equipment to be determined, and standardize and generate the historical weight of each fault type; The real-time score of each fault type and the historical weight of each fault type are weighted to generate a total score of each fault type, and the equipment fault is determined based on the total score of each fault type; According to the degree of production impact matched by equipment failure, the operating status of the injection molding equipment is remotely controlled and a fault alarm is issued according to the degree of production impact.

Images