Industrial internet terminal equipment self-adaptive identification method based on incremental learning

Abstract

The invention discloses an industrial internet terminal equipment adaptive identification method based on incremental learning, and the method comprises the steps: carrying out the updating of equipment health state incremental learning parameters in a vibration spectrum of industrial internet terminal equipment, obtaining an equipment health state incremental learning result, and outputting an equipment health state classification threshold value; performing feature extraction in the current spectrum of the industrial internet terminal equipment to obtain an overload time period spectrum, and performing adaptive identification to obtain an overload risk coefficient; and an equipment overload distinguishing result is obtained through classification in combination with the overload risk coefficient, and when a safety peak value is exceeded, a danger instruction is sent to an equipment supervision engineer interaction end. According to the invention, the precision of equipment overload discrimination can be improved, the overload state can be handled in time, and safe and stable operation of equipment is ensured.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial internet terminal equipment overload identification, and particularly relates to an industrial internet terminal equipment adaptive identification method based on incremental learning. BACKGROUND

[0002] With the rapid development of industrial internet, the demand for digitalization and intelligentization of industrial equipment is increasing. Traditional equipment diagnosis and maintenance methods often rely on manual experience and periodic inspection, which is difficult to meet the requirements of real-time monitoring and efficient diagnosis in modern industrial complex environments. The adaptive identification technology of industrial internet terminal equipment emerges as the times require, which combines big data, Internet of Things and artificial intelligence technology to monitor the running state of equipment in real time and automatically identify the health state and failure mode of equipment. Among them, the adaptive identification method based on incremental learning can continuously absorb new information during equipment operation, dynamically adjust the model, and ensure that the equipment maintains efficient and stable operation under changing working conditions. The advantage of this technology is that it can continuously learn without restarting the model, reducing the cost of data retraining, and improving the accuracy of fault prediction and equipment optimization.

[0003] Although traditional equipment monitoring and diagnosis technology has been widely used in the industrial field, there are still many limitations. The existing industrial internet terminal equipment overload monitoring method cannot accurately identify and judge the overload state of the industrial internet terminal equipment, resulting in insufficient accuracy of equipment overload identification, and causing the technical problem of poor reliability and safety of equipment operation. SUMMARY

[0004] In order to overcome the shortcomings and deficiencies of the prior art, the present application provides an industrial internet terminal equipment adaptive identification method based on incremental learning.

[0005] In a first aspect, the present application provides an industrial internet terminal equipment adaptive identification method based on incremental learning, which comprises: Step S1, using an acceleration sensor and a current transformer, collecting the frequency spectrum of the industrial internet terminal equipment within a limited range, obtaining the vibration frequency spectrum of the industrial internet terminal equipment and the current frequency spectrum of the industrial internet terminal equipment, collecting the equipment operation risk coefficient in unit time, and obtaining the incremental learning parameter configuration of the industrial internet terminal equipment by feature clustering; Step S2, using the incremental learning parameter configuration of the industrial internet terminal equipment, updating the incremental learning parameters of the health state of the industrial internet terminal equipment in the vibration frequency spectrum of the industrial internet terminal equipment, obtaining the incremental learning result of the health state of the industrial internet terminal equipment, and obtaining the health state classification threshold of the industrial internet terminal equipment; Step S3, according to the industrial internet terminal equipment health state classification threshold, the overload period spectrum is obtained by feature extraction in the industrial internet terminal equipment current spectrum, the overload type adaptive identification is carried out, and the overload risk coefficient is obtained; Step S4, according to the industrial internet terminal equipment health state incremental learning result, the internet terminal equipment fault risk coefficient is calculated, the overload identification result of the industrial internet terminal equipment in the limited range is obtained by combining the overload risk coefficient, and the terminal equipment overload identification result is compared with the preset safety peak value. When the safety peak value is exceeded, a danger instruction is sent to the equipment supervision engineer interactive terminal.

[0006] Secondly, the industrial internet terminal equipment adaptive identification method based on incremental learning is realized by different units, and the units include: The industrial internet terminal equipment frequency spectrum integration unit is used to collect the frequency spectrum of the industrial internet terminal equipment in the limited range by using the acceleration sensor and the current transformer, and obtain the industrial internet terminal equipment vibration frequency spectrum and the industrial internet terminal equipment current frequency spectrum. The device incremental learning parameter configuration acquisition unit is used to collect the device operation risk coefficient in unit time, and obtain the industrial internet terminal equipment incremental learning parameter configuration by feature clustering. The device health state incremental learning parameter updating unit is used to update the industrial internet terminal equipment health state incremental learning parameter in the industrial internet terminal equipment vibration frequency spectrum by using the industrial internet terminal equipment incremental learning parameter configuration, obtain the industrial internet terminal equipment health state incremental learning result, and obtain the industrial internet terminal equipment health state classification threshold. The overload type adaptive identification unit is used to obtain the overload period spectrum by feature extraction in the industrial internet terminal equipment current spectrum according to the industrial internet terminal equipment health state classification threshold, carry out the overload type adaptive identification, and obtain the overload risk coefficient. The overload identification result obtaining unit is used to obtain the industrial internet terminal equipment overload identification result in the limited range by combining the overload risk coefficient according to the industrial internet terminal equipment health state incremental learning result, calculate the internet terminal equipment fault risk coefficient, and compare the terminal equipment overload identification result with the preset safety peak value. When the safety peak value is exceeded, a danger instruction is sent to the equipment supervision engineer interactive terminal.

[0007] Beneficial effects: This invention proposes an adaptive identification method for industrial internet terminal devices based on incremental learning. Utilizing the incremental learning parameters configured for the industrial internet terminal devices, the method updates the health status parameters of the industrial internet terminal devices within the vibration spectrum of the devices, obtaining the incremental learning results of the health status and deriving a health status classification threshold. Based on the health status classification threshold, the method extracts features from the current spectrum of the industrial internet terminal devices to obtain the spectrum of overload periods, performs adaptive identification of overload types, and obtains an overload risk coefficient. Based on the incremental learning results of the health status, the method calculates the fault risk coefficient of the internet terminal devices. Combining this with the overload risk coefficient, the method categorizes the overload identification results of industrial internet terminal devices within a defined range. The overload identification results are compared with a preset safety peak value. When the safety peak value is exceeded, a danger command is sent to the equipment monitoring engineer's interactive terminal. This invention improves the accuracy of overload identification of industrial internet terminal devices through the incremental learning process, enabling timely and effective measures to address overload conditions. This invention ensures the safe operation of equipment and accurately processes the fault risk coefficient, achieving the technical effect of ensuring the safe and stable operation of equipment. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0009] Figure 1 This is a flowchart illustrating the overall steps of the method of the present invention; Figure 2 This is a diagram showing the composition of the method operation unit of the present invention. Detailed Implementation

[0010] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of this application can be combined with each other. The following describes the application in further detail with reference to the accompanying drawings and specific embodiments.

[0011] like Figure 1 As shown, this application provides an adaptive identification method for industrial internet terminal devices based on incremental learning, the method specifically including the following steps: Step S1, using an acceleration sensor and a current transformer, collecting the frequency spectrum of the industrial internet terminal device within a limited range, obtaining the vibration frequency spectrum of the industrial internet terminal device and the current frequency spectrum of the industrial internet terminal device, collecting the equipment operation risk coefficient in unit time, and obtaining the industrial internet terminal device incremental learning parameter configuration through feature clustering; Specifically, the data collection stage includes data collection by the acceleration sensor and the current transformer.

[0012] The acceleration sensor collects the vibration frequency spectrum. The goal is to monitor the vibration of the equipment through the acceleration sensor, obtain the mechanical vibration signal, and identify the running state and fault risk of the equipment. The process is to install the acceleration sensor on the key components (such as bearings, shafts, and housings) to collect vibration signals in real time. Through Fourier transform, the vibration signal is converted from the time domain to the frequency domain to generate the vibration frequency spectrum. Data: The vibration frequency spectrum data points contain multiple frequency components and their corresponding amplitude information.

[0013] The current transformer collects the current frequency spectrum. The goal is to monitor the current change of the equipment through the current transformer, capture the electrical characteristics, and help identify electrical faults (such as motor rotor imbalance, winding faults, etc.). The process is to install the current transformer on the power line of the equipment to collect the current signal of the equipment in real time. Through Fourier transform, the current signal is analyzed to obtain the current frequency spectrum. Data: The current frequency spectrum data also contains multiple frequency components and their corresponding current amplitude information.

[0014] The risk coefficient of the equipment operation in unit time is collected. Risk coefficient calculation: based on the real-time monitoring data of the vibration frequency spectrum and the current frequency spectrum, the running state of the equipment is analyzed, and the risk coefficient in unit time is calculated.

[0015] The risk coefficient can be based on frequency outliers, amplitude over-limit values, and energy spectrum distribution parameters. Common methods include monitoring key frequency components (such as normal working frequency of the equipment, harmonics, bearing characteristic frequency, etc.) and determining whether they deviate from the normal value.

[0016] The calculation formula is:

[0017] Wherein, represents a specific feature value of the vibration or current frequency spectrum, is the mean value of historical data, is the standard deviation, is the number of frequency spectrum features.

[0018] The risk coefficient can dynamically change over time, indicating the potential fault probability of the equipment under different running states.

[0019] Feature clustering, goal: Cluster the collected vibration spectrum and current spectrum data through clustering algorithms to identify and classify device operation modes. Process: Feature extraction: Extract key features in the spectrum, such as vibration and current peaks at specific frequencies, harmonic components, etc., as feature vectors.

[0020] Clustering algorithm: Apply K-Means, DBSCAN or other clustering algorithms to group different spectral features. Clustering results can distinguish different operating states of the device, such as normal operation state, potential fault state, etc. Results: Each cluster represents a device operating state, and related spectral features are clustered into the same category. Cluster centers can be used to represent the typical feature values of this state.

[0021] Incremental learning parameter configuration, goal: Obtain parameter configurations for incremental learning based on feature clustering results to help the model continuously adapt to new data changes. Process: According to the features and risk coefficients in the clustering results, select appropriate incremental learning model parameters such as learning rate, forgetting rate, model update frequency, etc.

[0022] These parameters determine the response speed and model adaptation ability of the incremental learning model when processing new device data. For example: Learning rate: Determines the magnitude of weight updates when new data arrives. A higher learning rate is suitable for rapidly changing device states.

[0023] Forgetting rate: Controls the retention of old knowledge. If the device's operating state changes significantly, a higher forgetting rate helps the model quickly adapt to the new state.

[0024] Update frequency: Determines how often the model is updated. Frequent updates are suitable for scenarios where device states change frequently.

[0025] Automatic parameter tuning: Dynamically adjust the parameters of incremental learning using real-time clustering results to enable the model to continuously learn and adapt to changes in device operating states.

[0026] Model update through incremental learning: Update the model based on real-time data streams and risk assessment to adapt to the current operating state of the device. The incremental learning model can continuously learn new data based on the features extracted in the previous step, and continuously optimize the device's state recognition and fault prediction capabilities.

[0027] Step S2, using the industrial internet terminal device incremental learning parameter configuration, updating the industrial internet terminal device health state incremental learning parameters in the industrial internet terminal device vibration spectrum, obtaining the industrial internet terminal device health state incremental learning result, and obtaining the industrial internet terminal device health state classification threshold; Specifically, the industrial internet terminal device increment learning parameter configuration is then used to sequentially update the industrial internet terminal device health state increment learning parameters within the vibration spectrum of the industrial internet terminal device, and the industrial internet terminal device increment learning parameter with the largest increment learning batch size is selected as the industrial internet terminal device health state increment learning result; the industrial internet terminal device health state classification threshold is obtained, Classification threshold determination: During the increment learning process, the model automatically adjusts the classification boundary according to the changes in the vibration spectrum, thereby determining the classification threshold of each health state.

[0028] These thresholds represent the dividing points of different health states, which can be defined by frequency characteristics and amplitude characteristics: Normal and warning state threshold: When the characteristic frequency and amplitude in the vibration spectrum exceed this threshold, the device state enters the warning state from the normal state.

[0029] Warning and failure state threshold: When the vibration characteristics in the spectrum further intensify and exceed the failure threshold, the device state is marked as a failure state.

[0030] Threshold adjustment: During the increment learning process, as new data is introduced, the classification threshold may also be dynamically adjusted to adapt the model to long-term changes in device operating conditions.

[0031] Classification threshold expression: Frequency domain classification threshold: By analyzing specific frequencies (such as bearing failure frequencies, rotational frequencies, etc.), a threshold is set. When the amplitude of these frequency components in the spectrum exceeds the predetermined threshold, it indicates that the device has entered a failure state.

[0032] Energy threshold: By monitoring the total energy in the spectrum or the energy level within a specific frequency range, an energy threshold is set. When the vibration energy of the device is too high, it indicates a potential failure.

[0033] Step S3, according to the industrial internet terminal device health state classification threshold, the overload period spectrum is obtained by feature extraction in the current spectrum of the industrial internet terminal device, and the overload type is adaptively identified, and the overload risk coefficient is obtained; Specifically, according to the industrial internet terminal device health state classification threshold, the frequency spectrum feature extraction is performed in the industrial internet terminal device current frequency spectrum, that is, the industrial internet terminal device health state classification threshold is processed in the industrial internet terminal device current frequency spectrum, and the frequency spectrum outside the industrial internet terminal device health state classification threshold, that is, the overload period spectrum, is obtained. Then, the overload type adaptive identification is performed on the overload period spectrum, wherein the overload type includes overload type, overload thickness and other characteristic information, and the overload type includes mechanical overload, electrical overload, thermal overload, hydraulic overload and vibration overload, and an overload risk coefficient is obtained.

[0034] By performing feature extraction in the industrial internet terminal device current frequency spectrum based on the industrial internet terminal device health state classification threshold, the accuracy of frequency spectrum feature extraction can be improved, thereby improving the accuracy of the overload period spectrum.

[0035] Step S4, according to the industrial internet terminal device health state incremental learning result, the internet terminal device fault risk coefficient is calculated, combined with the overload risk coefficient, the industrial internet terminal device overload identification result in the limited range is obtained, and the terminal device overload identification result is compared with the preset safety peak value. When the safety peak value is exceeded, a danger instruction is sent to the equipment supervision engineer interactive terminal.

[0036] Specifically, according to the industrial internet terminal device health state incremental learning result, the industrial internet terminal device incremental learning parameter corresponding to the industrial internet terminal device fault risk coefficient in the industrial internet terminal device health state incremental learning result is obtained, and then the overload environment correlation is comprehensively analyzed according to the industrial internet terminal device fault risk coefficient and the overload risk coefficient. The overload environment correlation in the limited range is obtained. Then, it is judged whether the overload environment correlation is greater than the preset environment correlation threshold, and the preset environment correlation threshold can be set by the person skilled in the art according to the actual situation. When the safety peak value is exceeded, a danger instruction is sent to the equipment supervision engineer interactive terminal.

[0037] The industrial internet terminal equipment adaptive recognition method based on incremental learning can solve the technical problems of the existing industrial internet terminal equipment overload monitoring method. Due to the inability to accurately adaptively identify and judge the overload state of the industrial internet terminal equipment, the accuracy of equipment overload identification is insufficient, resulting in poor reliability and safety of equipment operation. First, the acceleration sensor and the current transformer are used to collect the frequency spectrum of the industrial internet terminal equipment within a limited range, and the vibration frequency spectrum of the industrial internet terminal equipment and the current frequency spectrum of the industrial internet terminal equipment are obtained. Then, the equipment operation risk coefficient in unit time is collected, and the feature clustering obtains the incremental learning parameter configuration of the industrial internet terminal equipment; using the incremental learning parameter configuration of the industrial internet terminal equipment, the incremental learning parameter update of the industrial internet terminal equipment health state is carried out in the vibration frequency spectrum of the industrial internet terminal equipment, the incremental learning result of the industrial internet terminal equipment health state is obtained, and the health state classification threshold of the industrial internet terminal equipment is obtained. According to the health state classification threshold of the industrial internet terminal equipment, the overload period spectrum is obtained by feature extraction in the current frequency spectrum of the industrial internet terminal equipment, the overload type adaptive identification is carried out, and the overload risk coefficient is obtained. Finally, according to the incremental learning result of the industrial internet terminal equipment health state, the fault risk coefficient of the internet terminal equipment is calculated, combined with the overload risk coefficient, the overload identification result of the industrial internet terminal equipment within the limited range is obtained, and the terminal equipment overload identification result is compared with the preset safety peak value. When it exceeds the safety peak value, a danger instruction is sent to the equipment supervision engineer interaction terminal. By collecting the equipment operation risk coefficient to match the incremental learning parameter configuration of the industrial internet terminal equipment, then according to the incremental learning parameter configuration of the industrial internet terminal equipment, the incremental learning parameter update of the industrial internet terminal equipment health state is carried out in the vibration frequency spectrum of the industrial internet terminal equipment, the incremental learning result of the industrial internet terminal equipment health state and the health state classification threshold of the industrial internet terminal equipment are obtained; further according to the health state classification threshold of the industrial internet terminal equipment, the overload period spectrum is obtained by feature extraction in the current frequency spectrum of the industrial internet terminal equipment, and the overload type adaptive identification is carried out on the overload period spectrum, and the overload risk coefficient is obtained. On the other hand, based on the incremental learning result of the industrial internet terminal equipment health state, the fault risk coefficient of the industrial internet terminal equipment is determined; finally, based on the fault risk coefficient of the industrial internet terminal equipment and the overload risk coefficient, the industrial internet terminal equipment overload identification result is determined by comprehensive analysis, if the industrial internet terminal equipment overload identification result meets the alarm rule, the alarm is triggered. The accuracy of the industrial internet terminal equipment overload identification can be improved, so that effective measures can be taken in time to deal with the overload state of the equipment, and the technical effect of ensuring the safe and stable operation of the equipment is achieved.

[0038] The equipment operation risk coefficient in a unit time is collected, and the feature clustering is used to obtain the incremental learning parameter configuration of the industrial internet terminal equipment, specifically including: The equipment operation risk coefficient in a unit time is collected; according to the historical industrial internet terminal equipment monitoring data set, the equipment operation risk coefficients of different capacity demand devices are obtained, and the industrial internet terminal equipment model list under the equipment operation risk coefficient of each capacity demand device is obtained; The feature clustering relationship between the different capacity demand equipment operation risk coefficients and the different capacity demand industrial internet terminal equipment model list is established; According to the equipment operation risk coefficient, the feature clustering is performed to obtain the incremental learning parameter configuration of the industrial internet terminal equipment.

[0039] Specifically, first, the equipment operation risk coefficient in a unit time is collected through a multi-modal sensor, wherein the multi-source sensor can be set according to a preset equipment operation characteristic index, and the preset equipment operation characteristic index can be set according to actual conditions; then, based on the historical industrial internet terminal equipment monitoring data set, the equipment operation risk coefficients of different capacity demand devices are obtained, and the industrial internet terminal equipment model list under the equipment operation risk coefficient of each capacity demand device is obtained, wherein in addition to being affected by the equipment operation risk coefficient, the capacity of the industrial internet terminal equipment is also affected by other factors. Further, the feature clustering relationship between the different capacity demand equipment operation risk coefficients and the different capacity demand industrial internet terminal equipment model list is established, and the equipment operation risk coefficient is input into the feature clustering relationship for matching to obtain the incremental learning parameter configuration of the industrial internet terminal equipment.

[0040] Further, the incremental learning parameter configuration of the industrial internet terminal equipment is used to update the incremental learning parameter of the health state of the industrial internet terminal equipment in the vibration frequency spectrum of the industrial internet terminal equipment, specifically including: In the incremental learning parameter configuration of the industrial internet terminal equipment, the initial incremental learning parameter of the industrial internet terminal equipment is selected according to the equipment operation power, the incremental learning parameter of the health state of the industrial internet terminal equipment is updated in the vibration frequency spectrum of the industrial internet terminal equipment, until the preset accuracy is reached, the initial incremental learning result is obtained, and the initial incremental learning result includes the initial result incremental learning batch size; Specifically, first, an initial industrial internet terminal device incremental learning parameter is selected in the industrial internet terminal device incremental learning parameter configuration according to the equipment running power, the initial industrial internet terminal device incremental learning parameter is any one of the industrial internet terminal device incremental learning parameters in the industrial internet terminal device incremental learning parameter configuration, and according to the initial industrial internet terminal device incremental learning parameter, the industrial internet terminal device health state is updated in the industrial internet terminal device vibration spectrum, until a preset update number is reached, and the maximum incremental learning batch size in unit time is output as the initial incremental learning result, wherein the initial incremental learning result includes an initial result incremental learning batch size, and the preset update number can be set according to the required accuracy.

[0041] Further, the present application also includes the following steps: An initial industrial internet terminal device incremental learning parameter is selected in the industrial internet terminal device incremental learning parameter configuration according to the equipment running power, and the industrial internet terminal device health state is updated in the industrial internet terminal device vibration spectrum according to the equipment running power, to obtain an initial incremental learning result. The data size corresponding to all extreme points in the industrial internet terminal device vibration spectrum is obtained, and an initial incremental learning batch size is obtained. The initial industrial internet terminal device incremental learning parameter is continuously used to update the industrial internet terminal device health state in the industrial internet terminal device vibration spectrum according to the equipment running power, to obtain a dynamic incremental learning result, and a dynamic incremental learning batch size is calculated and obtained. The incremental learning parameter updating is continuously performed until a preset accuracy is reached, and the maximum incremental learning batch size and the corresponding incremental learning result are output as the initial result incremental learning batch size and the initial incremental learning result.

[0042] Specifically, first, an initial industrial internet terminal device incremental learning parameter is selected in the industrial internet terminal device incremental learning parameter configuration according to the equipment running power, and according to the initial industrial internet terminal device incremental learning parameter, the industrial internet terminal device health state is updated in the industrial internet terminal device vibration spectrum according to the equipment running power, that is, the data size corresponding to the extreme points in the industrial internet terminal device vibration spectrum is obtained, to obtain an initial incremental learning result. Further, the data size corresponding to all extreme points in the industrial internet terminal device vibration spectrum, that is, the ratio of the extreme points of the initial incremental learning result to the number of all extreme points in the industrial internet terminal device vibration spectrum, is calculated, and the ratio is set as an initial incremental learning batch size.

[0043] Then continue to use the initial industrial internet terminal equipment incremental learning parameters, and the industrial internet terminal equipment vibration spectrum is used to incrementally learn the health status of the industrial internet terminal equipment according to the equipment running power, obtain a dynamic incremental learning result, and calculate a dynamic incremental learning batch size; continue to update the incremental learning parameters until the preset incremental learning number is reached, then output the maximum incremental learning batch size and the corresponding incremental learning result, take the maximum incremental learning batch size as the initial result incremental learning batch size, and take the incremental learning result as the initial incremental learning result.

[0044] According to the initial result incremental learning batch size, an initial incremental learning forgetting rate is calculated and obtained; According to the equipment running power, the dynamic industrial internet terminal equipment incremental learning parameters are selected in the industrial internet terminal equipment incremental learning parameter configuration, and the learning rate of the dynamic industrial internet terminal equipment incremental learning parameters is calculated and obtained in combination with the initial incremental learning forgetting rate; Further, the present application further comprises the following steps: According to the equipment running power, the dynamic industrial internet terminal equipment incremental learning parameters are selected in the industrial internet terminal equipment incremental learning parameter configuration; The difference between the dynamic industrial internet terminal equipment incremental learning parameters and the initial industrial internet terminal equipment incremental learning parameters is adaptively recognized, and a unit time covariance is calculated and obtained; The convolution result of the unit time covariance and the initial incremental learning forgetting rate is calculated and obtained as the learning rate.

[0045] According to the learning rate, the dynamic industrial internet terminal equipment incremental learning parameters are used to perform incremental learning parameter update of the health status of the industrial internet terminal equipment in the industrial internet terminal equipment vibration spectrum; According to the equipment running power, the industrial internet terminal equipment incremental learning parameters are selected to perform incremental learning parameter update of the health status of the industrial internet terminal equipment, until the preset accuracy rate is reached, the optimal incremental learning batch size corresponding to the industrial internet terminal equipment incremental learning parameters is output, and the industrial internet terminal equipment incremental learning result is taken as the health status of the industrial internet terminal equipment; According to the industrial internet terminal equipment health status incremental learning result, an internet terminal equipment health status classification threshold is generated and calculated.

[0046] Specifically, the learning rate is judged according to a preset learning rate threshold. If the learning rate is greater than or equal to the preset learning rate threshold, the dynamic industrial internet terminal device incremental learning parameters are used to update the incremental learning parameters of the industrial internet terminal device health state in the industrial internet terminal device vibration spectrum. If the learning rate is less than the preset learning rate threshold, the dynamic industrial internet terminal device incremental learning parameters are reselected, and the learning rate is recalculated until the industrial internet terminal device incremental learning parameters with a learning rate greater than or equal to the preset learning rate threshold are obtained.

[0047] According to the equipment running power, the industrial internet terminal device incremental learning parameters are selected to update the incremental learning parameters of the industrial internet terminal device health state until the preset incremental learning parameter update times are met. The preset incremental learning parameter update times can be set based on demand, and the optimal incremental learning batch size corresponding to the industrial internet terminal device incremental learning parameters is output, and the industrial internet terminal device incremental learning parameters are used as the incremental learning result of the industrial internet terminal device health state. Then, according to the incremental learning result of the industrial internet terminal device health state, the incremental learning result of the industrial internet terminal device health state is extracted to obtain the industrial internet terminal device health state classification threshold.

[0048] Further, according to the industrial internet terminal device health state classification threshold, the overload period spectrum is obtained by feature extraction in the industrial internet terminal device current spectrum, and the overload type adaptive identification is performed to obtain the overload risk coefficient. Step four of the present application includes: The time nodes of the industrial internet terminal device vibration spectrum and the industrial internet terminal device current spectrum are overlapped. According to the industrial internet terminal device health state classification threshold, the overload period spectrum is obtained by feature extraction in the industrial internet terminal device current spectrum. According to the industrial internet terminal device overload identification data record, the unit time overload period spectrum curve and the unit time overload risk coefficient configuration are obtained. Using the unit time overload period spectrum curve and the unit time overload risk coefficient configuration, based on the incremental K-Means model, the overload type adaptive identification parameters are established, the feature adaptive identification is performed on the overload period spectrum, and the overload risk coefficient is obtained.

[0049] Specifically, the industrial internet terminal device vibration frequency spectrum and the industrial internet terminal device current frequency spectrum are overlapped in time nodes, that is, the industrial internet terminal device vibration frequency spectrum and the industrial internet terminal device current frequency spectrum are corresponded in industrial internet terminal device positions; further, according to the industrial internet terminal device health state classification threshold, the frequency spectrum feature is extracted in the industrial internet terminal device current frequency spectrum, and the frequency spectrum except the industrial internet terminal device health state classification threshold is set as the overload period spectrum. The industrial internet terminal device overload discrimination data record is called to obtain the unit time overload period spectrum curve and the unit time overload risk coefficient configuration.

[0050] The incremental K-Means model has strong ability in frequency spectrum feature extraction, which is mainly composed of convolution layer, pooling layer and full connection layer, and extracts features in the frequency spectrum through layer-by-layer stacking. Based on the incremental K-Means model, an overload category adaptive recognition parameter is established, which is a neural network model that can be updated and optimized in machine learning. The input data of the overload category adaptive recognition parameter is the overload period spectrum, and the output data is the overload risk coefficient. Further, the unit time overload period spectrum curve and the unit time overload risk coefficient configuration are used as training data sets to update and supervise the learning of the overload category adaptive recognition parameter until the overload category adaptive recognition parameter that meets the expected training constraints is obtained. Finally, the overload category adaptive recognition parameter is used to perform feature adaptive recognition on the overload period spectrum to obtain the overload risk coefficient. By establishing the overload category adaptive recognition parameter based on the incremental K-Means model for feature adaptive recognition, the accuracy, reliability and efficiency of the overload risk coefficient can be improved.

[0051] Further, according to the incremental learning result of the industrial internet terminal device health state, the internet terminal device failure risk coefficient is calculated, and the overload risk coefficient is combined to obtain the industrial internet terminal device overload discrimination result in the limited range. Step five of the present application includes: Obtain the industrial internet terminal device failure risk coefficient corresponding to the industrial internet terminal device incremental learning parameter in the incremental learning result of the industrial internet terminal device health state as the basic failure risk coefficient; According to the optimal incremental learning batch size, the error of the basic failure risk coefficient is calculated to obtain the error failure risk coefficient; Based on the overload discrimination data of the industrial internet terminal device, the unit time failure risk coefficient configuration, the unit time overload risk coefficient configuration and the unit time overload environment correlation list are obtained; The unit time fault risk coefficient configuration, the unit time overload risk coefficient configuration, and the unit time overload environment correlation list are used to perform high-dimensional data processing on the error fault risk coefficient and the overload risk coefficient based on a support vector machine model, and an overload environment correlation is obtained as the overload discrimination result of the industrial internet terminal device.

[0052] Specifically, an industrial internet terminal device fault risk coefficient corresponding to an industrial internet terminal device incremental learning parameter in the industrial internet terminal device health state incremental learning result is obtained, and the industrial internet terminal device fault risk coefficient is taken as a basic fault risk coefficient. Then, the basic fault risk coefficient is calculated according to the optimal incremental learning batch size, that is, the basic fault risk coefficient is divided by the optimal incremental learning batch size, and the calculation result is taken as an error fault risk coefficient, which is used to eliminate the device fault risk coefficient deviation caused by the incremental learning deviation, so that the error fault risk coefficient is more accurate.

[0053] The overload discrimination data of the industrial internet terminal device is called, and the overload discrimination data is historical overload discrimination data record of the industrial internet terminal device. According to the overload discrimination data, a unit time fault risk coefficient configuration, a unit time overload risk coefficient configuration, and a unit time overload environment correlation list are obtained, wherein the unit time fault risk coefficient, the unit time overload risk coefficient, and the unit time overload environment correlation correspond one by one.

[0054] Further based on the principle of the support vector machine model, the unit time fault risk coefficient and the unit time overload risk coefficient are taken as kernel function inputs, the corresponding unit time overload environment correlation is taken as the boundary relaxation variable, the unit time fault risk coefficient configuration, the unit time overload risk coefficient configuration, and the unit time overload environment correlation list are taken as establishment data, and the overload environment correlation is taken as the overload discrimination result of the industrial internet terminal device. The accuracy and efficiency of the overload environment correlation can be improved based on the support vector machine model, thereby improving the accuracy of the overload discrimination.

[0055] In summary, the industrial internet terminal device adaptive recognition method based on incremental learning provided in the present application has the following technical effects: The application improves the accuracy of overload identification of industrial internet terminal equipment, so that effective measures can be taken in time to deal with the overload state of the equipment, and the technical effect of ensuring the safe and stable operation of the equipment is achieved. By calculating the difference value of the incremental learning parameters of the industrial internet terminal equipment, and setting the learning rate based on the difference value deviation, the invalid incremental learning can be reduced, and the incremental learning efficiency of the model can be improved. By using the support vector machine model, the accuracy and efficiency of the overload environment correlation can be improved, thereby improving the accuracy of overload identification.

[0056] As shown in Figure 2 The industrial internet terminal equipment adaptive identification method based on incremental learning is implemented by different units, including: The industrial internet terminal equipment spectrum integration unit is used to collect the spectrum of the industrial internet terminal equipment within a limited range by using an acceleration sensor and a current transformer, to obtain the vibration spectrum of the industrial internet terminal equipment and the current spectrum of the industrial internet terminal equipment; The device incremental learning parameter configuration acquisition unit is used to collect the equipment operation risk coefficient within a unit time, and obtain the incremental learning parameter configuration of the industrial internet terminal equipment by feature clustering; The device health state incremental learning parameter updating unit is used to update the health state incremental learning parameters of the industrial internet terminal equipment in the vibration spectrum of the industrial internet terminal equipment by using the incremental learning parameter configuration of the industrial internet terminal equipment, to obtain the health state incremental learning result of the industrial internet terminal equipment, and to obtain the health state classification threshold of the industrial internet terminal equipment; The overload type adaptive identification unit is used to obtain the overload period spectrum by feature extraction in the current spectrum of the industrial internet terminal equipment according to the health state classification threshold of the industrial internet terminal equipment, to perform adaptive identification of the overload type, and to obtain the overload risk coefficient; The overload identification result obtaining unit is used to calculate the internet terminal equipment fault risk coefficient according to the health state incremental learning result of the industrial internet terminal equipment, to obtain the overload identification result of the industrial internet terminal equipment within the limited range by classification in combination with the overload risk coefficient, and to compare the terminal equipment overload identification result with the preset safety peak value, and to send a danger instruction to the equipment supervision engineer interaction terminal when the safety peak value is exceeded.

[0057] Further, the device incremental learning parameter configuration acquisition unit is also used to: Collecting a device operation risk coefficient in a unit time; According to the historical industrial internet terminal device monitoring data set, the operation risk coefficient of different capacity demand devices is obtained, as well as the list of industrial internet terminal device models under the operation risk coefficient of each capacity demand device; Establishing the characteristic clustering relationship of the different capacity demand device operation risk coefficient and the different capacity demand industrial internet terminal device model list; According to the device operation risk coefficient, the characteristic clustering is carried out to obtain the industrial internet terminal device incremental learning parameter configuration.

[0058] Further, the device health state incremental learning parameter updating unit is also used for: Selecting initial industrial internet terminal device incremental learning parameters according to the device operation power in the industrial internet terminal device incremental learning parameter configuration, and updating the incremental learning parameters of the industrial internet terminal device health state in the industrial internet terminal device vibration spectrum until the preset accuracy is reached, obtaining the initial incremental learning result, which includes the initial result incremental learning batch size; According to the initial result, the incremental learning forgetting rate is obtained; Selecting dynamic industrial internet terminal device incremental learning parameters according to the device operation power in the industrial internet terminal device incremental learning parameter configuration, and combining the initial incremental learning forgetting rate to calculate the learning rate of the dynamic industrial internet terminal device incremental learning parameters; According to the learning rate, the dynamic industrial internet terminal device incremental learning parameters are used to update the incremental learning parameters of the industrial internet terminal device health state in the industrial internet terminal device vibration spectrum; Continue to select industrial internet terminal device incremental learning parameters according to the device operation power to update the incremental learning parameters of the industrial internet terminal device health state until the preset accuracy is reached, output the optimal incremental learning batch size corresponding to the industrial internet terminal device incremental learning parameters as the incremental learning result of the industrial internet terminal device health state; According to the incremental learning result of the industrial internet terminal device health state, the internet terminal device health state classification threshold is generated and calculated.

[0059] Further, the device health state incremental learning parameter updating unit is also used for: Selecting initial industrial internet terminal device incremental learning parameters according to the device operation power in the industrial internet terminal device incremental learning parameter configuration, and obtaining the initial incremental learning result according to the device operation power incremental learning of the industrial internet terminal device health state in the industrial internet terminal device vibration spectrum; Obtaining the data size of all extreme points in the vibration frequency spectrum of the industrial internet terminal device, obtaining an initial incremental learning batch size; Continuing to use the initial industrial internet terminal device incremental learning parameters to perform incremental learning on the industrial internet terminal device health status according to the device operating power in the vibration frequency spectrum of the industrial internet terminal device, obtaining a dynamic incremental learning result, and calculating a dynamic incremental learning batch size; Continuing to update the incremental learning parameters until a preset accuracy is reached, outputting the maximum incremental learning batch size and the corresponding incremental learning result as the initial result incremental learning batch size and the initial incremental learning result.

[0060] Further, the device health state incremental learning parameter updating unit is also used for: Selecting dynamic industrial internet terminal device incremental learning parameters according to the device operating power in the industrial internet terminal device incremental learning parameter configuration; Adaptively identifying the difference between the dynamic industrial internet terminal device incremental learning parameters and the initial industrial internet terminal device incremental learning parameters, and calculating a unit time covariance; Calculating the convolution result of the unit time covariance and the initial incremental learning forgetting rate as the learning rate.

[0061] Further, the overload category adaptive identification unit is also used for: Overlapping time nodes of the industrial internet terminal device vibration frequency spectrum and the industrial internet terminal device current frequency spectrum; According to the industrial internet terminal device health state classification threshold, obtaining an overload period spectrum by feature extraction in the industrial internet terminal device current frequency spectrum; According to the industrial internet terminal device overload identification data record, obtaining a unit time overload period spectrum curve and a unit time overload risk coefficient configuration; Using the unit time overload period spectrum curve and the unit time overload risk coefficient configuration, establishing an overload category adaptive identification parameter based on an incremental K-Means model, performing feature adaptive identification on the overload period spectrum, and obtaining an overload risk coefficient.

[0062] Further, the overload identification result obtaining unit is also used for: Obtaining an industrial internet terminal device fault risk coefficient corresponding to the industrial internet terminal device incremental learning parameter in the industrial internet terminal device health state incremental learning result as a basic fault risk coefficient; According to the optimal incremental learning batch size, performing error calculation on the basic fault risk coefficient to obtain an error fault risk coefficient; Based on the overload discrimination data of the industrial internet terminal equipment, obtain a unit time fault risk coefficient configuration, a unit time overload risk coefficient configuration and a unit time overload environment correlation list; Using the unit time fault risk coefficient configuration, the unit time overload risk coefficient configuration and the unit time overload environment correlation list, based on a support vector machine model, high-dimensional data processing is performed on the error fault risk coefficient and the overload risk coefficient to obtain an overload environment correlation as the overload discrimination result of the industrial internet terminal equipment.

[0063] The above description of disclosed embodiments enables one of ordinary skill in the art to make and use the application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0064] Obviously, various modifications and changes can be made to the present application by those skilled in the art without departing from the spirit and scope of the application. Accordingly, the application intends to include all such modifications and changes insofar as they fall within the scope of the application and its equivalents.

Claims

1. An adaptive identification method for industrial internet terminal devices based on incremental learning, characterized in that, The method includes: Step S1: Using an accelerometer and a current transformer, collect the spectrum of industrial internet terminal devices within a limited range to obtain the vibration spectrum and current spectrum of industrial internet terminal devices, collect the equipment operation risk coefficient per unit time, and obtain the incremental learning parameter configuration of industrial internet terminal devices through feature clustering. The formula for calculating the equipment operation risk factor is: in, This represents specific characteristic values ​​of the vibration or current spectrum. It is the average of historical data. It is the standard deviation. It is the number of spectral features; Step S2: Using the incremental learning parameter configuration of the industrial internet terminal device, update the incremental learning parameters of the health status of the industrial internet terminal device within the vibration spectrum of the industrial internet terminal device, obtain the incremental learning result of the health status of the industrial internet terminal device, and obtain the classification threshold of the health status of the industrial internet terminal device. Step S3: Based on the health status classification threshold of the industrial internet terminal equipment, extract the overload period spectrum from the current spectrum of the industrial internet terminal equipment, perform adaptive identification of overload types, and obtain the overload risk coefficient. Step S4: Based on the incremental learning results of the health status of the industrial internet terminal equipment, calculate the fault risk coefficient of the internet terminal equipment. Combined with the overload risk coefficient, classify and obtain the overload identification results of the industrial internet terminal equipment within the limited range. Compare the overload identification results of the terminal equipment with the preset safety peak. When the safety peak is exceeded, send a danger command to the interactive terminal of the equipment supervision engineer.

2. The adaptive identification method for industrial internet terminal devices based on incremental learning according to claim 1, characterized in that, Step S1 includes: Collect the equipment operation risk coefficient per unit time, and based on the historical industrial internet terminal equipment monitoring data set, obtain the equipment operation risk coefficient for different capacity requirements, as well as the list of industrial internet terminal equipment models per unit time under each capacity requirement equipment operation risk coefficient. Establish the feature clustering relationship between the operational risk coefficients of equipment with different capacity requirements and the list of industrial internet terminal equipment models with different capacity requirements; Based on the equipment operation risk coefficient, feature clustering is performed to obtain the incremental learning parameter configuration of the industrial internet terminal equipment.

3. The adaptive identification method for industrial internet terminal devices based on incremental learning according to claim 1, characterized in that, Step S2 includes: Within the configuration of incremental learning parameters for industrial internet terminal devices, initial incremental learning parameters for industrial internet terminal devices are selected based on the device's operating power. Within the vibration spectrum of the industrial internet terminal devices, incremental learning parameters for the health status of the industrial internet terminal devices are updated until a preset accuracy rate is reached, thereby obtaining initial incremental learning results. The initial incremental learning results include the initial result incremental learning batch size. Based on the initial incremental learning batch size, the initial incremental learning forgetting rate is calculated. Within the incremental learning parameter configuration of the industrial internet terminal device, dynamic industrial internet terminal device incremental learning parameters are selected according to the device operating power. Combined with the initial incremental learning forgetting rate, the learning rate of the dynamic industrial internet terminal device incremental learning parameters is calculated. Based on the learning rate, the incremental learning parameters of the dynamic industrial internet terminal device are used to update the health status of the industrial internet terminal device within the vibration spectrum of the industrial internet terminal device. Based on the equipment operating power, select the incremental learning parameters of the industrial internet terminal equipment to update the incremental learning parameters of the health status of the industrial internet terminal equipment until the preset accuracy is reached. Then, output the incremental learning parameters of the industrial internet terminal equipment corresponding to the optimal incremental learning batch size as the incremental learning result of the health status of the industrial internet terminal equipment. Based on the incremental learning results of the health status of the industrial internet terminal devices, a classification threshold for the health status of internet terminal devices is generated and calculated.

4. The adaptive identification method for industrial internet terminal devices based on incremental learning according to claim 3, characterized in that, Within the incremental learning parameter configuration of the industrial internet terminal device, initial incremental learning parameters are selected based on the device's operating power. The incremental learning parameters for the health status of the industrial internet terminal device are updated within the vibration spectrum of the industrial internet terminal device, including: Within the configuration of incremental learning parameters for industrial internet terminal devices, initial incremental learning parameters for industrial internet terminal devices are selected based on the device's operating power. The health status of industrial internet terminal devices is incrementally learned based on the device's operating power using the vibration spectrum of the industrial internet terminal devices to obtain initial incremental learning results. Obtain the data scale corresponding to all extreme points in the vibration spectrum of the industrial internet terminal equipment, and obtain the initial incremental learning batch size; Continuing to utilize the initial incremental learning parameters of the industrial internet terminal equipment, incremental learning of the health status of the industrial internet terminal equipment based on the equipment operating power is performed on the vibration spectrum of the industrial internet terminal equipment to obtain dynamic incremental learning results, and the dynamic incremental learning batch size is calculated. Continue updating the incremental learning parameters until the preset accuracy is reached, and output the maximum incremental learning batch size and the corresponding incremental learning result as the initial result incremental learning batch size and initial incremental learning result.

5. The adaptive identification method for industrial internet terminal devices based on incremental learning according to claim 3, characterized in that, Within the industrial internet terminal device incremental learning parameter configuration, dynamic industrial internet terminal device incremental learning parameters are selected based on the device's operating power. Combined with the initial incremental learning forgetting rate, the learning rate of the dynamic industrial internet terminal device incremental learning parameters is calculated, including: Within the configuration of incremental learning parameters for industrial internet terminal devices, dynamic incremental learning parameters for industrial internet terminal devices are selected based on the device's operating power. Adaptively identify the difference between the incremental learning parameters of the dynamic industrial internet terminal device and the initial incremental learning parameters of the industrial internet terminal device, and calculate the unit time covariance. The convolution result of the unit-time covariance and the initial incremental learning-forgetting rate is used as the learning rate.

6. The adaptive identification method for industrial internet terminal devices based on incremental learning according to claim 1, characterized in that, Step S3 includes: Time nodes are overlapped between the vibration spectrum and current spectrum of the industrial internet terminal device. Based on the health status classification threshold of the industrial internet terminal equipment, the spectrum of the overload period is obtained by feature extraction within the current spectrum of the industrial internet terminal equipment. Based on the overload identification data records of industrial internet terminal devices, obtain the spectrum curve of overload period per unit time and the configuration of overload risk coefficient per unit time; Using the overload period spectrum curve and overload risk coefficient configuration, an adaptive identification parameter for overload type is established based on the incremental K-Means model. The overload period spectrum is then subjected to feature adaptive identification to obtain the overload risk coefficient.

7. The adaptive identification method for industrial internet terminal devices based on incremental learning according to claim 1, characterized in that, Step S4 includes: Obtain the industrial internet terminal device fault risk coefficient corresponding to the industrial internet terminal device incremental learning parameter in the industrial internet terminal device health status incremental learning result, and use it as the basic fault risk coefficient. Based on the optimal incremental learning batch size, the error is calculated on the basic fault risk coefficient to obtain the error fault risk coefficient. Based on the overload identification data of industrial internet terminal equipment, obtain the unit time fault risk coefficient configuration, the unit time overload risk coefficient configuration, and the unit time overload environment correlation list; Using the unit-time fault risk coefficient configuration, unit-time overload risk coefficient configuration, and unit-time overload environment correlation list, based on the support vector machine model, high-dimensional data processing is performed on the error fault risk coefficient and overload risk coefficient to obtain the overload environment correlation, which serves as the overload identification result of the industrial internet terminal equipment.

8. The adaptive identification method for industrial internet terminal devices based on incremental learning according to any one of claims 1-7, characterized in that, This method is implemented through different units, including: An industrial internet terminal device spectrum integration unit is used to collect the spectrum of industrial internet terminal devices within a defined range using an accelerometer and a current transformer, and to obtain the vibration spectrum and current spectrum of industrial internet terminal devices. The device incremental learning parameter configuration acquisition unit is used to collect the device operation risk coefficient within a unit of time and obtain the incremental learning parameter configuration of the industrial Internet terminal device through feature clustering. The device health status incremental learning parameter update unit is used to update the device health status incremental learning parameter update within the vibration spectrum of the industrial internet terminal device using the configuration of the incremental learning parameter of the industrial internet terminal device, to obtain the incremental learning result of the industrial internet terminal device health status, and to obtain the classification threshold of the industrial internet terminal device health status. An overload type adaptive identification unit is used to extract the overload period spectrum from the current spectrum of the industrial internet terminal equipment based on the health status classification threshold of the industrial internet terminal equipment, perform overload type adaptive identification, and obtain the overload risk coefficient. The overload identification result acquisition unit is used to calculate the fault risk coefficient of the Internet terminal equipment based on the incremental learning result of the health status of the Industrial Internet terminal equipment, and combine the overload risk coefficient to classify the overload identification results of the Industrial Internet terminal equipment within the limited range. The overload identification results of the terminal equipment are compared with the preset safety peak. When the safety peak is exceeded, a danger command is sent to the interactive terminal of the equipment supervision engineer.

Images