Industrial telephone set with remote monitoring function and system

Abstract

The invention relates to the technical field of industrial telephones, in particular to an industrial telephone system with a remote monitoring function. The telephone set system comprises a telephone set main body, a main control module which is responsible for coordinating and managing the work of each module, a monitoring module which is used for collecting each operation parameter of the industrial telephone set in real time, a communication module which supports a plurality of communication modes, a power supply module which provides a working power supply for the industrial telephone set, and an industrial internet interface. An intelligent collaboration module integrating AI / ML, a data storage and transmission module, a display and operation module, a main control module, a monitoring module, a communication module, a power supply module and an industrial internet interface are integrated in a telephone set main body, and all the modules are connected through a high-speed bus or an interface; the intelligent cooperation module comprises an AI / ML acceleration unit, and the AI / ML acceleration unit is used for helping a user make a decision and understand the decision.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial telephones, and in particular to an industrial telephone with a remote monitoring function. Background Art

[0002] In industrial production, telephones are crucial for ensuring smooth communication between personnel, while industrial control systems (ICs) are the core of automated production. However, traditional industrial telephones are independent of ICSs, with limited integration. Telephones cannot directly access real-time data from ICSs, such as equipment operating status and production parameters. Simultaneously, the ICSs cannot access operating status information from the telephones. This leads to severe information silos, hindering comprehensive monitoring and coordinated management of the production process. Furthermore, due to a lack of effective integration, coordinated control between telephones and ICSs is impossible. When the ICS detects a device failure, it cannot automatically notify maintenance personnel via telephone, nor can maintenance personnel remotely operate or control the equipment directly via telephone. This low level of integration hinders the intelligentization of industrial production and emergency response capabilities.

[0003] In the case of industrial telephone fault diagnosis, traditional industrial telephone systems with remote monitoring capabilities can provide relatively accurate fault diagnosis results. However, their black-box nature makes it difficult for users to understand how the model makes decisions. This lack of transparency not only reduces user trust in the model but also limits its widespread application in real-world industrial environments.

[0004] Therefore, an industrial telephone and system with a high degree of integration and a remote monitoring function are proposed. Summary of the Invention

[0005] The purpose of the present invention is to provide an industrial telephone with a remote monitoring function: to solve the problem that although the industrial telephone system with a remote monitoring function in the prior art can provide fault diagnosis results with a certain accuracy, its black box characteristics make it difficult for users to understand how the model makes decisions.

[0006] The purpose of the present invention can be achieved through the following technical solutions:

[0007] An industrial telephone system with a remote monitoring function includes: a telephone body, a main control module responsible for coordinating and managing the work of various modules, a monitoring module for real-time collection of various operating parameters of the industrial telephone, a communication module supporting multiple communication methods, a power module for providing working power for the industrial telephone, an industrial Internet interface, an intelligent collaboration module integrating AI / ML, a data storage and transmission module, and a display and operation module. The main control module, monitoring module, communication module, power module, industrial Internet interface, intelligent collaboration module, data storage and transmission module, and display and operation module are integrated into the telephone body, and the modules are connected via a high-speed bus or interface. The intelligent collaboration module includes an AI / ML acceleration unit, which is used to help users make and understand decisions.

[0008] Preferably, the AI / ML intelligent collaboration module is implemented as follows:

[0009] First, the multimodal data generated by industrial telephones is preprocessed, including cleaning, standardization, and time series alignment. Then, the data is fused and feature weighting is performed using the attention mechanism. High-order fault features are extracted by combining topological data analysis. A graph neural network model is constructed for training and diagnosis. During the diagnosis process, the SHAP method is used to generate an interpretable report. Finally, based on the fault diagnosis accuracy, the fault diagnosis accuracy is and recall Quantitatively evaluate and optimize model performance.

[0010] Preferably, the multimodal data includes equipment operating status, production parameters and telephone call records, and the data are fused through canonical correlation analysis.

[0011] Preferably, after the data are fused through canonical correlation analysis, a time series alignment operation is also performed on the fused data to transform the static analysis of the multimodal data into a dynamic analysis of the multimodal data.

[0012] Preferably, the aligned time series data are used for similarity comparison and the Euclidean distance between the time series is calculated.

[0013] Preferably, a graph neural network model is constructed for training and diagnosis. The model is trained using a training set, and the model parameters are updated through a back propagation algorithm. The loss function is defined as a cross entropy loss function. ;

[0014] in, is the true label, is the model prediction probability, is the number of categories.

[0015] Preferably, use the test set to evaluate the fault diagnosis accuracy of the model , in, is the total number of samples, is the sample index number, Indicates the model The prediction results of samples, Indicates the model True labels or actual fault categories of samples, It is a true function that returns 1 if the condition inside the indicated function is true; otherwise it returns 0. This is a real example. It's a false counterexample. The value range is [0,1], and the closer to 1, the better the classification performance.

[0016] The advantages of the present invention are: Through canonical correlation analysis, sensor data, call records and log files are integrated, breaking through the information limitations of a single data source and making full use of the cross-modal correlation information of the device's operating status.

[0017] It solves the time axis offset problem caused by sampling rate differences in industrial telephone time series data, improves the comparability of failure modes of different devices and batches, and provides a high-quality time benchmark for subsequent feature extraction.

[0018] Multimodal data collected from industrial control systems and telephone monitoring modules, including equipment operating status, production parameters, and telephone call logs, is fused through canonical correlation analysis (CCA) data and coordinated with the Mapper algorithm to achieve full process optimization from state perception to root cause analysis. Sensor data captures the physical state of the equipment in real time, log files record operational history and process context, and call logs reflect human factors and communication reliability. The complementary nature of these three enables fault diagnosis to break through the limitations of a single data source. Using canonical correlation analysis (CCA) to fuse multimodal data, topological data analysis to extract high-order features, and combined with SHAP interpretation optimization, this significantly improves fault detection sensitivity, root cause analysis accuracy, and predictive maintenance capabilities. This data fusion model upgrades industrial telephones from simple communication terminals to intelligent sensing nodes, directly reducing maintenance costs, improving production continuity, and forming a closed loop from passive response to active intervention. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0020] Figure 1 This is a flowchart of an industrial telephone with remote monitoring function and a system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] The present invention provides an industrial telephone system with a remote monitoring function, the telephone system comprising: Telephone body; The main control module is responsible for coordinating and managing the work of each module. As the core control unit of the industrial telephone, the main control module can process data from each module in real time and make corresponding control decisions.

[0023] A monitoring module used to collect various operating parameters of industrial telephones in real time, including call status, signal strength, device temperature, power supply voltage and current. The monitoring module contains multiple sensors and detection circuits, including temperature sensors, signal strength detection circuits, voltage detection circuits and current detection circuits, for accurate and real-time acquisition of telephone operating status information.

[0024] The communication module supports multiple communication methods, including Ethernet, Wi-Fi, and 4G / 5G. The phone can select the appropriate communication method based on the actual industrial environment. The communication module is responsible for transmitting the operating parameters collected by the monitoring module to the remote monitoring center and supports data exchange with the Industrial Internet interface. The communication module has a dedicated data acquisition interface that connects to the industrial control system and the phone, ensuring that the industrial control system can obtain real-time equipment operating data, production parameters, and the phone's operating status information.

[0025] The power module provides working power for industrial telephones. The power module adopts a wide voltage input design to adapt to the large voltage fluctuations in industrial environments. The power module has overvoltage, overcurrent, and overtemperature protection functions. When an abnormal situation occurs, it can automatically cut off the power supply to protect the safety of the telephone.

[0026] The Industrial Internet Interface (III) acts as a bridge between industrial phones and the Industrial Internet platform, enabling communication protocol conversion and data exchange between the two. It supports multiple Industrial Internet communication protocols, including MQTT, CoAP, and HTTP / 2, and can connect to a wide range of Industrial Internet platforms. Industrial phones access the Industrial Internet platform through the III to obtain industrial data from the platform, including equipment operating data, production plans, and quality information. They also upload their operating parameters and status information to the platform, enabling data aggregation and sharing.

[0027] The intelligent collaboration module integrates AI / ML. This module performs real-time analysis of data from industrial control systems and telephones, enabling intelligent fault diagnosis and predictive maintenance. The intelligent collaboration module includes an AI / ML acceleration unit. This unit uses a low-power, high-performance AI acceleration chip as its core processor, such as the Google Edge TPU, Huawei Ascend 310, or NVIDIA Jetson, to run AI / ML algorithms. This unit helps users make and understand decisions, increasing trust in the telephone system and promoting its application in real-world industrial environments.

[0028] The data storage and transmission module is a high-speed, large-capacity storage device that can use an SSD or NVMe hard drive to store historical phone data and data of the models used to run the intelligent collaboration module that integrates AI / ML.

[0029] The display and operation module is equipped with a touch screen and multi-function operation buttons on the phone body. The touch screen is used to display the operating status, operating parameters, and fault diagnosis results of the industrial phone.

[0030] The main control module, monitoring module, communication module, power module, industrial Internet interface, intelligent collaboration module, data storage and transmission module, and display and operation module are integrated into the phone body. Each module is connected via a high-speed bus or interface. Trained AI / ML models are loaded onto the AI ​​accelerator chip, enabling real-time data analysis and intelligent decision-making.

[0031] Reference Figure 1 The intelligent collaborative module integrating AI / ML in this application can also generate explanations of industrial telephone fault diagnosis results, which has important practical significance. Its implementation method is as follows.

[0032] Step 1: Data Collection: The AI / ML acceleration unit collects historical data from the telephone monitoring module. This historical data includes equipment operating status (including temperature, pressure, vibration, etc.), production parameters (including output and efficiency, etc.), and telephone call records (including call duration and signal strength, etc.). This can be understood as the AI / ML acceleration unit collecting three types of data from the industrial telephone system: sensor data, log files, and call records.

[0033] Based on this, the AI / ML acceleration unit builds a data matrix ,

[0034]

[0035]

[0036]

[0037] in, is the sensor data matrix, is the log file matrix, is the call record matrix, is the sample size, 、 、 The number of features of each mode is used to distinguish the data of different modes. The data are synchronized in time. For each sample, the corresponding timestamp is recorded. , .

[0038] And perform data cleaning, including removing noise, processing missing values, standardization / normalization. The above data cleaning methods are all common processing methods and belong to the basic knowledge of those skilled in the art, so they will not be elaborated here. Still expressed in original parameters.

[0039] Step 2: Data fusion. Use canonical correlation analysis to fuse multimodal data and calculate the covariance matrix of sensor data and call records respectively. , and their cross-covariance matrices ;

[0040] Solving for generalized eigenvalues ​​to obtain canonical correlation variables and , and the canonical correlation coefficient , here are typical correlation variables associated with the sensor data matrix, are the canonical correlation variables associated with the call record matrix. They represent the linear combination with the highest correlation between the two sets of data. In the canonical correlation analysis CCA, the variables The solution process is obtained by maximizing the sensor data matrix and the call record matrix. The Lagrange multiplier method can be used to construct the Lagrange function and calculate its partial derivatives. We will not go into details here.

[0041] Get related variables and Then select the first k pairs of typical correlation variables and construct the fused feature matrix .

[0042]

[0043] In the industrial telephone fault diagnosis scenario, the AI / ML acceleration unit uses canonical correlation analysis to fuse multimodal data. CCA maximizes the correlations between sensor data, call records, and log files to construct a structured correlation matrix for the three modal data. For example, a strong correlation can be found between abnormal bearing temperature (sensor) and missing lubrication operation records (log); and the temporal synchronization between sudden signal strength drops (call records) and electromagnetic interference events (logs) can be found. This structured correlation provides a traceable causal path for diagnostic results.

[0044] Step 3: Time series alignment, for the fused feature matrix , construct the distance matrix ,in , , and are the eigenvalues ​​of the i-th and j-th time points of the two time series, respectively, and the cumulative distance matrix is ​​initialized ,in , , , backtracking the optimal alignment path, from Start until .

[0045] Industrial field data often suffers from timeline misalignment due to sampling rate differences and network latency. For example, sensor data is sampled at 1Hz, log files are recorded based on event triggers, and call records are stored based on call start timestamps. When these misalignments occur, an abnormal temperature peak (from the sensor) may be mistakenly associated with a lubrication operation (from the log) in a later time period, leading to incorrect attribution. Time series alignment maps the timeline to a unified benchmark, ensuring strict temporal correspondence between the three events: temperature anomaly, lubrication operation, and call interruption.

[0046] Aligned time series facilitate the discovery of temporal causality, improve optimization model training, and enhance the ability to learn temporal patterns. Through time series alignment, the industrial telephone system has transitioned from "static snapshot analysis" to "dynamic process tracing," aligning the decision-making process of the AI / ML acceleration unit with the temporal causal thinking model of industrial field engineers.

[0047] Step 4: Similarity comparison: use the aligned time series data to perform similarity comparison and calculate the Euclidean distance between the time series. ,in , s is the covariance matrix. By comparing the Euclidean distance between real-time data and historical normal state data, anomalies that deviate significantly from the baseline can be quickly identified.

[0048] Data from industrial sites often comes from a variety of sensors, logging systems, and communication devices. These data sources have varying sampling rates, timestamp accuracy, or time bases. For example, sensor data may be collected at a 1Hz frequency, while log files may only record data when specific events occur. This inconsistency in the temporal dimension causes data points to shift along the time axis. Direct similarity comparisons may mistakenly interpret data from different time points as similar or correlated. Time series alignment eliminates this time shift, ensuring that the compared data points correspond in time. This allows similarity comparisons to more accurately reflect the true relationship between data, without causing spurious correlations or differences due to time shifts.

[0049] Step 5: Calculate attention weights and feature weights to construct the attention weight matrix ,in is the number of features after fusion, and the softmax function is used to normalize the attention weights. , the fused feature matrix and the attention weight matrix Multiply to get the weighted feature matrix .

[0050]

[0051] After calculating the attention weights and feature weights, the AI / ML acceleration unit dynamically assigns the importance of different modalities and features, enabling it to focus on data dimensions that have a greater impact on diagnostic results, such as temperature anomalies in sensor data and sudden drops in signal strength in call records, thereby improving diagnostic accuracy. At the same time, the attention weight distribution provides engineers with an explainable decision path.

[0052] Step 6: Topological data analysis, using the Mapper algorithm to analyze the weighted feature matrix Perform topological data analysis, select a set of covering functions, divide the data space into multiple overlapping regions, calculate local topological features on each covering tree region, use the persistent homology algorithm to calculate the 0-dimensional and 1-dimensional persistent homology of each region, and combine the local topological features on all covering tree regions into a global topological representation.

[0053] Topological data analysis reveals the underlying structural patterns in high-dimensional data by extracting topological features such as clusters, holes, and loops. In industrial telephone fault diagnosis, it can identify complex fault modes that are difficult to capture with traditional methods, enhance the sensitivity of diagnostic models to abnormal data, and provide interpretable topological fingerprints to help engineers understand the spatial distribution characteristics of faults.

[0054] Step 7: Deep learning model training. Select graph convolutional network as the deep learning model, build GCN model, input the feature representation obtained by topological data analysis, and define the propagation rules of GCN layer. .

[0055]

[0056] in , is the adjacency matrix with self-connection added, yes The degree matrix of It is The feature matrix of the layer, It is The weight matrix of the layer, is the activation function, and the model output is the fault diagnosis result, such as fault type and fault location.

[0057] Step 6: Model training and optimization. Divide the dataset into training, validation, and test sets, with proportions of 70%, 15%, and 15%, respectively.

[0058] Use the training set to train the model, update the model parameters through the back propagation algorithm, and define the loss function as the cross entropy loss function .

[0059]

[0060] is the true label, is the model prediction probability, is the number of categories.

[0061] Then use Adam optimizer to update parameters;

[0062] and are the first-order moment estimate and the second-order moment estimate, is the learning rate, is a small constant.

[0063] After model training, the validation set is used to select the model and tune the hyperparameters. The number of GCN layers, learning rate, batch size and other hyperparameter combinations are increased to find the optimal model.

[0064] Step 7: SHAP value calculation explanation and explanation generation. For each sample and model prediction result, use the SHAP method to calculate the Shapley value of each feature and select a feature subset. , calculate the marginal contribution of this subset to the model prediction results ;

[0065] For all possible feature subsets Perform weighted summation to obtain features Shapley value ;

[0066] in The Shapley value is a collection of all features. It generates an explanation based on the Shapley value, which shows the contribution of each feature to the fault diagnosis result. A bar chart is used to display the Shapley value of each feature. A positive value indicates that the feature increases the probability of failure, while a negative value indicates that the feature decreases the probability of failure.

[0067] Step 8, explanation optimization, collects user satisfaction and feedback on the explanations through questionnaires, user interviews, etc., adjusts the explanation generation process based on user feedback, and adjusts the parameters of the SHAP method to achieve better explanation results.

[0068] Step 9: Fault diagnosis optimization, use the test set to evaluate the fault diagnosis accuracy of the model , recall rate ;

[0069]

[0070] in, is the total number of samples, is the sample index number, Indicates the model The prediction results of samples, Indicates the model True labels or actual fault categories of samples, It is a true function that returns 1 if the condition inside the indicated function is true; otherwise it returns 0. This is a real example. is a false counterexample. Evaluate the accuracy, stability, and consistency of the explanation. Compare the explanation generated by the explanation generator with the actual cause of the failure and calculate the accuracy of the explanation. A core indicator that reflects the overall classification capability of the model. Its value range is [0,1]. The closer to 1, the better the classification performance. In industrial scenarios, high accuracy means fewer missed detections and false positives, which are directly related to equipment maintenance costs and production safety. Measures the model's ability to identify fault samples, with particular attention paid to the missed detection rate. In the field of industrial safety, a high recall rate means a lower risk of missed fault detection and is a key indicator for ensuring production continuity.

[0071] Through the composition and implementation details of the above-mentioned industrial telephones, the intelligent collaborative industrial telephones based on the industrial Internet platform can realize real-time analysis of industrial control systems and telephone data, provide intelligent fault diagnosis and predictive maintenance functions, thereby significantly improving the intelligence level, collaborative efficiency and emergency response capabilities of industrial production.

[0072] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0073] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. An industrial telephone system with remote monitoring function, characterized in that: The telephone system includes: a telephone body, a main control module responsible for coordinating and managing the work of each module, a monitoring module for real-time collection of various operating parameters of the industrial telephone, a communication module that supports multiple communication methods, a power module that provides working power for the industrial telephone, an industrial Internet interface, an intelligent collaboration module integrated with AI / ML, a data storage and transmission module, and a display and operation module. The main control module, monitoring module, communication module, power module, industrial Internet interface, intelligent collaboration module, data storage and transmission module, and display and operation module are integrated in the telephone body, and the modules are connected via a high-speed bus or interface. The intelligent collaboration module includes an AI / ML acceleration unit, which is used to help users make and understand decisions.

2. The method according to claim 1, characterized in that The AI / ML intelligent collaboration module is implemented as follows: First, the multimodal data generated by industrial telephones is preprocessed, including cleaning, normalization, and time-series alignment. The data is then fused and feature weighted using an attention mechanism. High-level fault features are extracted using topological data analysis. Build graph neural network models for training and diagnosis; During the diagnosis process, the SHAP method is used to generate an interpretable report; finally, based on the fault diagnosis accuracy and recall Quantitatively evaluate and optimize model performance.

3. The method according to claim 2, characterized in that The multimodal data includes equipment operating status, production parameters and telephone call records, and the data is fused through canonical correlation analysis.

4. The method according to claim 3, characterized in that After fusing the data through canonical correlation analysis, time series alignment is performed on the fused data to transform the static analysis of multimodal data into dynamic analysis of multimodal data.

5. The method according to claim 4, characterized in that The aligned time series data are used for similarity comparison and the Euclidean distance between the time series is calculated.

6. The method according to claim 2, characterized in that Construct a graph neural network model for training and diagnosis. Use the training set to train the model, update the model parameters through the back propagation algorithm, and define the loss function as the cross entropy loss function. ; in, is the true label, is the model prediction probability, is the number of categories.

7. The method according to claim 2, characterized in that Use the test set to evaluate the fault diagnosis accuracy of the model , in, is the total number of samples, is the sample index number, Indicates the model The prediction results of samples, Indicates the model True labels or actual fault categories of samples, It is a true function that returns 1 if the condition inside the indicated function is true; otherwise it returns 0. It is a real example. It's a false counterexample. The value range is [0,1], and the closer to 1, the better the classification performance.

Images