Industrial Internet of Things data classification method and device and storage medium

Abstract

The invention relates to the technical field of industrial data processing, in particular to an industrial Internet of Things data classification method and device and a storage medium, and the method comprises the steps: collecting sensor data in real time, and carrying out the filtering processing; performing time-frequency domain feature extraction on the sensor data to obtain data window features; analyzing a window comprehensive fluctuation index according to the sensor data, and performing type marking on the sensor data according to the window comprehensive fluctuation index; constructing a window feature vector based on the data window feature, and analyzing the type similarity based on the window feature vector and the type tag of the sensor data; constructing a data classification model based on the data window features and the type tag of the sensor data to obtain a data classification tag; and triggering a data classification model updating rule according to the type similarity and the type mark of the sensor data. According to the invention, accurate classification analysis of the Internet of Things sensor data is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial data processing, and in particular, to an industrial Internet of Things data classification method, device, and storage medium. Background Art

[0002] As one of the core technologies of Industry 4.0, the industrial Internet of Things realizes real-time monitoring and intelligent decision-making of the production process by connecting devices, sensors, and data analysis platforms. In industrial scenarios, sensor data is a key representation of the health status of devices, and its efficient classification and anomaly detection are crucial for predictive maintenance, fault warning, and production optimization. However, with the increasing complexity of industrial equipment and the explosive growth of data volume, traditional data processing methods face severe challenges in terms of real-time performance, classification accuracy, and adaptability.

[0003] In the prior art, there are many disadvantages in the classification of industrial Internet of Things data: insufficient suppression of data noise: industrial environment noise leads to a low signal-to-noise ratio of sensor signals; single-dimensional feature extraction: The prior art mostly relies on a single sensor or simple time-domain features, ignoring the time-frequency domain collaborative analysis of multi-modal data; rigid classification rules and poor adaptability: traditional methods use fixed thresholds or static classification rules and cannot adapt to the dynamic working conditions of devices; contradiction between computing resources and real-time performance: although high-complexity algorithms can improve feature accuracy, they consume a large amount of edge computing resources and are difficult to meet the millisecond-level real-time response requirements; weak fault tracing ability: most methods can only judge the "existence of anomalies", but cannot accurately locate the fault type through multi-dimensional feature association. Summary of the Invention

[0004] The purpose of the present invention is to provide an industrial Internet of Things data classification method, device, and storage medium to solve at least one of the problems existing in the prior art.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions: An industrial Internet of Things data classification method includes: Real-time collecting sensor data and performing filtering processing; Performing time-frequency domain feature extraction on the sensor data to obtain data window features; Analyzing the comprehensive fluctuation index of the sensor data analysis window according to the sensor data, and marking the type of the sensor data according to the window comprehensive fluctuation index; Constructing a window feature vector based on the data window features, and analyzing the type similarity based on the window feature vector and the type marking of the sensor data; Constructing a data classification model based on the data window features and the type marking of the sensor data to obtain a data classification label; Trigger the data classification model update rule based on the type similarity and the type label of the sensor data.

[0006] Further, analyze the effective value of the first window based on the X-axis vibration velocity in the triaxial vibration sensor data, compare the effective value of the first window with the velocity threshold. If the effective value of the first window is greater than or equal to the velocity threshold, trigger the frequency spectrum refinement analysis; otherwise, do not trigger the frequency spectrum refinement analysis. The frequency spectrum refinement analysis is to perform a fast Fourier transform on the triaxial vibration sensor data to extract the rotation frequency and harmonic components of the monitoring target, and analyze the characteristics of the first data window based on the rotation frequency and harmonics. The expression of the characteristics of the first data window is: Rf1 = A2f + A3f / Af, where Rf1 represents the characteristics of the first data window, A2f represents the harmonic amplitude of twice the rotation frequency, A3f represents the harmonic amplitude of three times the rotation frequency, and Af represents the rotation frequency.

[0007] Further, analyze the characteristics of the second data window based on the infrared temperature sensor data, and take the ratio of the difference between the end temperature and the start temperature in the window to the window length as the characteristics of the second data window. Compare the acoustic emission energy of each data point in the acoustic emission sensor data with the energy threshold. If the acoustic emission energy is greater than or equal to the energy threshold, mark the current analyzed data point as an impact event; otherwise, do not mark the current analyzed data point. Analyze the signal phase based on the rotation frequency. The expression of the signal phase is θ = 2×π×Af, where θ represents the signal phase. Extract the impact events whose signal phase of the data points corresponding to the impact events belongs to [0, π / 2] as effective impact events, and take the ratio of the number of effective impact events to the number of impact events as the characteristics of the third data window.

[0008] Further, analyze the effective value of the second window based on the Y-axis vibration velocity in the triaxial vibration sensor data, analyze the effective value of the third window based on the Z-axis vibration velocity in the triaxial vibration sensor data, and perform dynamic weight allocation on the effective value of the first window, the effective value of the second window, the effective value of the third window, and the impact events within the window in M minutes to obtain the window comprehensive fluctuation index. Compare the window comprehensive fluctuation index with the index threshold to mark the sensor data as fluctuation data and steady-state data.

[0009] Further, construct a window feature vector based on the data window characteristics, and take the cosine similarity between the window feature vector of the sensor data marked as fluctuation data and the window feature vector of the sensor data marked as steady-state data as the type similarity.

[0010] Further, analyze the feature anomaly status based on the second data window feature and the third data window feature. If the second data window feature is greater than or equal to the second feature threshold, set the feature anomaly status as temperature mutation anomaly; otherwise, set the feature anomaly status as normal temperature. If the third data window feature is greater than or equal to the third feature threshold, set the feature anomaly status as acoustic emission damage anomaly; otherwise, set the feature anomaly status as normal acoustic emission. Construct a first-level data classification label based on the data window feature. When the first data window feature is greater than the first feature threshold, set the first-level data classification label as gear anomaly and trigger the second-level classification. When there is a VZ(i) greater than or equal to the kurtosis threshold, set the first-level data classification label as bearing anomaly and trigger the second-level classification. When NAf / N is greater than or equal to the balance threshold, set the first-level data classification label as running imbalance and trigger the second-level classification. Here, NAf represents the number of data points at 1 times the rotational frequency, N represents the number of data points within the window, and VZ(i) represents the vibration velocity of the Z-axis. The second-level classification is as follows: If there are at least two of gear anomaly, bearing anomaly, and running imbalance in the first-level data classification label, set the second-level data classification label as emergency warning. If the feature anomaly status is temperature mutation anomaly and acoustic emission damage anomaly, set the second-level data classification label as high-risk warning. If the feature anomaly status is temperature mutation anomaly or acoustic emission damage anomaly, set the second-level data classification label as low-risk warning. If the feature anomaly status is normal temperature and normal acoustic emission, set the second-level data classification label as normal.

[0011] Further, extract the sensor data within 1 hour. If the data points of the same sensor data are respectively marked as fluctuating data and steady-state data in the window comprehensive fluctuation index analysis of different M-minute windows, then regard the current analyzed data point as a drift data point, and use the ratio of the number of drift data points to the number of data points within 1 hour as the similar drift compensation parameter. If the similar drift compensation parameter is greater than or equal to the compensation threshold, update the analysis process of the type similarity.

[0012] Further, compare the type similarity with the similarity threshold. If the type similarity is less than or equal to the similarity threshold, trigger the data classification model update rule; otherwise, do not trigger the data classification model update rule. The triggered data classification model update rule is to reduce the first feature threshold, the second feature threshold, and the third feature threshold by G, where G represents the reduction ratio.

[0013] On the other hand, the present invention also provides an industrial Internet of Things data classification device, including: An acquisition and processing module for real-time acquisition of sensor data and filtering processing; A feature analysis module for extracting time-frequency domain features of the sensor data to obtain data window features; A type marking module for comprehensively analyzing the fluctuation index based on sensor data analysis windows and marking the types of sensor data according to the comprehensive window fluctuation index; A similarity analysis module for constructing a window feature vector based on data window features and analyzing type similarity based on the window feature vector and the type marking of sensor data; A data classification module for constructing a data classification model based on data window features and the type marking of sensor data to obtain data classification labels; A model update module for triggering a data classification model update rule based on type similarity and the type marking of sensor data.

[0014] On the other hand, the present invention also provides a storage medium storing instructions that, when run on a computer, cause the computer to execute the industrial Internet of Things data classification method described in any one of the above.

[0015] The beneficial effects of the present invention are as follows: By combining time-frequency domain feature extraction and dynamic weight allocation of multi-sensor data, the operating state of the device is comprehensively captured, effectively distinguishing noise from real anomalies. Through the comprehensive window fluctuation index and the model update rule, the system can dynamically adapt to changes in working conditions, avoiding misjudgment caused by data drift. By providing a differential response strategy through two-level classification labels, the rapid decision-making requirements of industrial scenarios are met; Using real-time filtering and 30-second window processing to ensure millisecond-level latency, suitable for high-speed production line monitoring; Triggered only when the effective value exceeds the threshold through spectrum refinement analysis, reducing redundant calculations, and through modular design analysis to support flexible expansion of new sensor types or algorithms, adapting to complex industrial Internet of Things ecosystems. Description of the Drawings

[0016] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for description in the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0017] Figure 1 It is a flowchart of the industrial Internet of Things data classification method in this embodiment.

[0018] Figure 2 It is a flowchart of the time-frequency domain feature extraction method in this embodiment.

[0019] Figure 3 It is a flowchart of the data classification model update method in this embodiment.

[0020] Figure 4 It is a structural schematic diagram of the industrial Internet of Things data classification device in this embodiment. Detailed Embodiments

[0021] To more clearly illustrate the present invention, the present invention will be further described below in conjunction with preferred embodiments and the accompanying drawings. Similar components in the drawings are denoted by the same reference numerals. Those skilled in the art should understand that the content specifically described below is illustrative rather than restrictive, and should not be used to limit the protection scope of the present invention.

[0022] It should be noted that although terms such as first, second, and third may be used in the embodiments of the present application for description, these descriptions should not be limited to these terms. These terms are only used to distinguish the descriptions. For example, without departing from the scope of the embodiments of the present application, the first may also be referred to as the second, and similarly, the second may also be referred to as the first.

[0023] Please refer to Figure 1 as shown, which is the industrial Internet of Things data classification method of this embodiment, including: Step S1, collecting sensor data in real time and performing filtering processing. The sensor data includes triaxial vibration sensor data, infrared temperature sensor data, and acoustic emission sensor data. The triaxial vibration sensor data is the vibration signal of the sensor monitoring target on the X, Y, and Z axes. The infrared temperature sensor data is the temperature signal of the sensor monitoring target. The acoustic emission sensor data is the acoustic emission signal of the sensor monitoring target. The acquisition method of the sensor data is obtained by importing data through the industrial Internet of Things management platform. In this embodiment, the monitoring target is a motor group. The triaxial vibration sensor data in this embodiment is the data reflecting the vibration magnitude value of the monitoring target, which can be represented by vibration displacement, vibration velocity, and vibration acceleration. In this embodiment, vibration velocity is used to represent the triaxial vibration sensor data, and its unit is mm / s.

[0024] Specifically, in step S1 of this embodiment, two-dimensional wavelet decomposition is used to process the triaxial vibration sensor data to eliminate environmental noise interference; the window length is set to 30 seconds, and median filtering is used to filter the infrared temperature sensor data to eliminate abnormal jump points in the infrared temperature sensor data; the acoustic emission sensor data with a frequency range of 50 kHz - 800 kHz is extracted to suppress low-frequency mechanical noise and extract effective impact event waveforms.

[0025] It can be understood that in this embodiment, the filtering processing method of the sensor data is not specifically limited, and those skilled in the art can freely set it. For example, methods such as moving average filtering and Kalman filtering can also be used to filter the sensor data, as long as the filtering processing of the data is satisfied.

[0026] Specifically, in step S1 of this embodiment, through two-dimensional wavelet decomposition, median filtering, and frequency band screening, environmental noise, abnormal jump points, and low-frequency interference in the data of the three-axis vibration, infrared temperature, and acoustic emission sensors are effectively eliminated, significantly improving the signal-to-noise ratio of the data. Real-time acquisition combined with window length optimization ensures the timeliness of data processing and provides high-quality input for subsequent feature extraction.

[0027] Please continue to refer to Figure 1 As shown, the industrial Internet of Things data classification method further includes: Step S2, performing time-frequency domain feature extraction on the sensor data to obtain data window features.

[0028] Specifically, in step S2 of this embodiment, through spectrum zoom analysis based on the three-axis vibration data, the rotation frequency and harmonic components are accurately extracted to enhance the sensitivity to mechanical fault features. By analyzing the temperature change rate feature of the infrared temperature data and screening the effective impact events of the acoustic emission signal, temperature mutation and mechanical damage can be respectively identified, improving the collaborative analysis ability of multi-modal data.

[0029] Please refer to Figure 2 As shown, it is a time-frequency domain feature extraction method, including: Step S21, analyzing the effective value of the first window of the three-axis vibration sensor data, and triggering spectrum zoom analysis based on the effective value of the first window to obtain the first data window feature.

[0030] Specifically, in step S21 of this embodiment, the effective value of the first window is analyzed based on the vibration velocity of the X axis in the three-axis vibration sensor data. The expression of the effective value of the first window is: , where R1 represents the effective value of the first window, VX(i) represents the vibration velocity of the X axis, i represents the data point number in the window, and N represents the number of data points in the window.

[0031] Specifically, in step S21 of this embodiment, the effective value of the first window is compared with the velocity threshold. If the effective value of the first window is greater than or equal to the velocity threshold, spectrum zoom analysis is triggered; otherwise, spectrum zoom analysis is not triggered. The spectrum zoom analysis is to perform a fast Fourier transform on the three-axis vibration sensor data to extract the rotation frequency and harmonic components of the monitoring target, and analyze the first data window feature based on the rotation frequency and harmonic. The expression of the first data window feature is: Rf1 = A2f + A3f / Af, where Rf1 represents the first data window feature, A2f represents the harmonic amplitude of twice the rotation frequency, A3f represents the harmonic amplitude of three times the rotation frequency, and Af represents the rotation frequency.

[0032] Specifically, in this embodiment, the speed threshold is set to 4 mm / s. In this embodiment, the value of the speed threshold is not specifically limited, and those skilled in the art can freely set it as long as it satisfies the comparison and analysis of the effective value of the first window. The value of the speed threshold should satisfy the range of [3 mm / s, 5 mm / s].

[0033] Please continue to refer to Figure 2 As shown, the time-frequency domain feature extraction method further includes: Step S22, analyzing the second data window feature based on the infrared temperature sensor data.

[0034] Specifically, in step S22 of this embodiment, when analyzing the second data window feature based on the infrared temperature sensor data, the ratio of the difference between the end temperature and the start temperature in the window to the window length is used as the second data window feature.

[0035] Specifically, in step S22 of this embodiment, when analyzing the second data window feature, the unit of the window length is converted to minutes so that the second data window feature represents the temperature change rate per minute.

[0036] Please continue to refer to Figure 2 As shown, the time-frequency domain feature extraction method further includes: Step S23, extracting the impact event and the signal phase based on the acoustic emission sensor data and the rotational frequency, and analyzing the third data window feature based on the impact event and the signal phase.

[0037] Specifically, in step S23 of this embodiment, the acoustic emission energy of each data point of the acoustic emission sensor data is compared with the energy threshold. If the acoustic emission energy is greater than or equal to the energy threshold, the current analyzed data point is marked as an impact event; otherwise, the current analyzed data point is not marked.

[0038] Specifically, in this embodiment, the energy threshold is set to 100 mV·ms. In this embodiment, the value of the energy threshold is not specifically limited, and those skilled in the art can freely set it as long as it satisfies the analysis of the impact event. The optimal value of the energy threshold should satisfy the range of [80 mV·ms, 120 mV·ms].

[0039] Specifically, in step S23 of this embodiment, the signal phase is analyzed according to the rotational frequency. The expression of the signal phase is θ = 2×π×Af, where θ represents the signal phase.

[0040] Specifically, in step S23 of this embodiment, the impact events with the signal phase of the data points corresponding to the impact events belonging to [0, π / 2] are extracted as effective impact events, and the ratio of the number of effective impact events to the number of impact events is used as the third data window feature.

[0041] Please continue to refer to Figure 1 As shown, the industrial Internet of Things data classification method further includes: Step S3: Analyze the comprehensive fluctuation index of the sensor data according to the sensor data analysis window, and mark the type of the sensor data according to the comprehensive fluctuation index of the window.

[0042] Specifically, in step S3 of this embodiment, the effective value of the second window is analyzed based on the Y-axis vibration velocity in the triaxial vibration sensor data, the effective value of the third window is analyzed based on the Z-axis vibration velocity in the triaxial vibration sensor data, and dynamic weight allocation is performed on the effective value of the first window, the effective value of the second window, the effective value of the third window, and the impact events within the window within M minutes to obtain the comprehensive fluctuation index of the window. The expression of the comprehensive fluctuation index of the window is: W = w1×σ(R1) / μ(R1) + w2×σ(R2) / μ(R2) + w3×σ(R3) / μ(R3) + w4×N1 / NM; where σ() represents calculating the standard deviation of the data in the parentheses, μ() represents calculating the average value of the data in the parentheses, R2 represents the effective value of the second window, R3 represents the effective value of the third window, w1 represents the first weight, w2 represents the second weight, w3 represents the third weight, w4 represents the fourth weight, w1 + w2 + w3 + w4 = 1, N1 represents the number of impact events within M minutes, NM represents the number of data points within M minutes, and M represents the comprehensive analysis parameter, 3 ≤ M ≤ 6.

[0043] Specifically, in this embodiment, the calculation methods of the effective value of the second window and the effective value of the third window are the same as those of the effective value of the first window, and no specific elaboration will be made in this embodiment.

[0044] Specifically, in this embodiment, the first weight is set to 0.4, the second weight is set to 0.2, the third weight is set to 0.2, and the fourth weight is set to 0.2. In this embodiment, no specific limitation is made on the values of each weight, and those skilled in the art can freely set them as long as the analysis of the comprehensive fluctuation index of the window is satisfied.

[0045] Specifically, in step S3 of this embodiment, the comprehensive fluctuation index of the window is compared with the index threshold. If the comprehensive fluctuation index of the window is greater than or equal to the index threshold, the sensor data corresponding to the comprehensive fluctuation index of the current analysis window is marked as fluctuating data; otherwise, the sensor data corresponding to the comprehensive fluctuation index of the current analysis window is marked as steady-state data.

[0046] Specifically, in this embodiment, the index threshold is set to 0.8. In this embodiment, no specific limitation is made on the value of the index threshold, and those skilled in the art can freely set it as long as the type marking of the sensor data is satisfied. The value of the index threshold should satisfy [0.7, 0.9].

[0047] Specifically, in step S3 of this embodiment, through dynamic weight allocation, a window comprehensive fluctuation index is generated by synthesizing multi-dimensional data, accurately distinguishing between fluctuation data and steady-state data, reducing the misjudgment rate, and using an exponential threshold comparison mechanism to achieve automatic marking of data types, providing a clear label basis for subsequent classification.

[0048] Please continue to refer to Figure 1 As shown, the industrial Internet of Things data classification method further includes: Step S4, constructing a window feature vector based on the data window feature, and analyzing the type similarity based on the window feature vector and the type marking of the sensor data.

[0049] Specifically, in step S4 of this embodiment, a window feature vector is constructed based on the data window feature, and the window feature vector is set as L, , and the cosine similarity between the window feature vector of the sensor data marked as fluctuation data and the window feature vector of the sensor data marked as steady-state data is used as the type similarity, where Rf2 represents the second data window feature and Rf3 represents the third data window feature.

[0050] It can be understood that in this embodiment, the analysis method of the type similarity is not specifically limited, and those skilled in the art can freely set it. For example, other methods such as click similarity and Euclidean similarity can also be set to analyze the type similarity.

[0051] Specifically, in step S4 of this embodiment, by constructing a multi-dimensional window feature vector and combining cosine similarity calculation, the correlation between fluctuation data and steady-state data is quantified, enhancing the interpretability and consistency of the classification model.

[0052] Please continue to refer to Figure 1 As shown, the industrial Internet of Things data classification method further includes: Step S5, constructing a data classification model based on the data window feature and the type marking of the sensor data to obtain a data classification label.

[0053] Specifically, in step S5 of this embodiment, a first-level data classification label is constructed according to the data window feature. When the first data window feature is greater than the first feature threshold, the first-level data classification label is set as gear abnormality and secondary classification is triggered; when there is VZ(i) greater than or equal to the kurtosis threshold, the first-level data classification label is set as bearing abnormality and secondary classification is triggered; when NAf / N is greater than or equal to the balance threshold, the first-level data classification label is set as running imbalance and secondary classification is triggered; where NAf represents the number of data points at 1 times the rotational frequency, and VZ(i) represents the vibration velocity of the Z axis.

[0054] Specifically, in this embodiment, the first feature threshold is set to 0.5, the kurtosis threshold is set to 5, and the balance threshold is set to 0.7. In this embodiment, no specific limitations are imposed on the above thresholds, and those skilled in the art can set them freely. The setting of the first feature threshold should satisfy belonging to [0.3, 0.6], the setting of the kurtosis threshold should satisfy belonging to [4, 6], and the setting of the balance threshold should satisfy belonging to [0.6, 0.7].

[0055] Specifically, in step S5 of this embodiment, when the data window feature is other than the above, the primary data classification label is set to normal.

[0056] Specifically, in step S5 of this embodiment, the feature anomaly state is analyzed according to the second data window feature and the third data window feature. If the second data window feature is greater than or equal to the second feature threshold, the feature anomaly state is set to temperature mutation anomaly; otherwise, the feature anomaly state is set to normal temperature. If the third data window feature is greater than or equal to the third feature threshold, the feature anomaly state is set to acoustic emission damage anomaly; otherwise, the feature anomaly state is set to normal acoustic emission.

[0057] Specifically, in this embodiment, the second feature threshold is set to 1, and the third feature threshold is set to 0.4. In this embodiment, no specific limitations are imposed on the above thresholds, and those skilled in the art can set them freely. The setting of the second feature threshold should satisfy belonging to [0.8, 1.2], and the setting of the third feature threshold should satisfy belonging to [0.3, 0.4].

[0058] Specifically, in step S5 of this embodiment, the secondary classification is as follows: if at least two of gear anomaly, bearing anomaly, and running imbalance exist simultaneously in the primary data classification label, the secondary data classification label is set to emergency warning; if the feature anomaly state is temperature mutation anomaly and acoustic emission damage anomaly, the secondary data classification label is set to high-risk warning; if the feature anomaly state is temperature mutation anomaly or acoustic emission damage anomaly, the secondary data classification label is set to low-risk warning; if the feature anomaly state is normal temperature and normal acoustic emission, the secondary data classification label is set to normal.

[0059] Specifically, in this embodiment, the secondary classification is triggered for analysis in the order of emergency warning, high-risk warning, low-risk warning, and normal in sequence.

[0060] Specifically, in step S5 of this embodiment, through a multi-level classification mechanism, the refined classification of faults is achieved to meet the differentiated warning requirements of industrial scenarios, and based on the automated trigger logic of threshold linkage, the classification efficiency is significantly improved, and manual intervention is reduced.

[0061] Please continue to refer to Figure 1 As shown, the industrial Internet of Things data classification method further includes: Step S6, trigger the data classification model update rule based on the type similarity and the type label of the sensor data.

[0062] Specifically, in step S6 of this embodiment, through the analysis of the similarity drift compensation parameter and the dynamic threshold adjustment, the model can adapt to the change of the data distribution, avoid the performance degradation caused by the environmental drift, use the type similarity threshold to control the update frequency, balance the model stability and real-time performance, and ensure the long-term reliability of the classification accuracy.

[0063] Please refer to Figure 3 as shown, which is a data classification model update method, including: Step S61, analyze the similarity drift compensation parameter according to the type label of the sensor data.

[0064] Specifically, in step S61 of this embodiment, extract the sensor data within 1 hour. If the data points of the same sensor data are respectively marked as fluctuating data and steady-state data in the window comprehensive fluctuation index analysis of different M-minute windows, then take the current analyzed data point as a drift data point, and take the ratio of the number of drift data points to the number of data points within 1 hour as the similarity drift compensation parameter.

[0065] Please continue to refer to Figure 3 as shown, the data classification model update method further includes: Step S62, update the analysis process of the type similarity according to the similarity drift compensation parameter.

[0066] Specifically, in step S62 of this embodiment, if the similarity drift compensation parameter is greater than or equal to the compensation threshold, update the analysis process of the type similarity, and set the updated type similarity as S1, S1 = S×e B ; otherwise, do not update the analysis process of the type similarity; where S1 represents the updated type similarity, S represents the type similarity before update, and B represents the similarity drift compensation parameter.

[0067] Specifically, in this embodiment, the compensation threshold is set to 0.03. In this embodiment, the value of the compensation threshold is not specifically limited, and those skilled in the art can freely set it as long as it satisfies the update of the type similarity. The setting of the compensation threshold should satisfy [0.02, 0.05].

[0068] Specifically, in this embodiment, only one judgment on whether to update the type similarity is performed in each type similarity analysis.

[0069] Please continue to refer to Figure 3 as shown, the data classification model update method further includes: Step S63: Trigger the data classification model update rule based on the type similarity.

[0070] Specifically, in step S63 of this embodiment, the type similarity is compared with the similarity threshold. If the type similarity is less than or equal to the similarity threshold, the data classification model update rule is triggered; otherwise, the data classification model update rule is not triggered. The triggered data classification model update rule is to reduce the first feature threshold, the second feature threshold, and the third feature threshold by G, where G represents the reduction ratio, and G = (1 - S) / 2 × 100%.

[0071] Specifically, the similarity threshold in this embodiment is set to 0.85. In this embodiment, the value of the similarity threshold is not specifically limited, and those skilled in the art can freely set it as long as it meets the trigger of the data classification model update rule. The setting of the similarity threshold should satisfy [0.8, 0.9].

[0072] Specifically, in this embodiment, when performing data operations, the unit is not considered, and only the numerical part of the data is used for operation and analysis.

[0073] Please refer to Figure 4 as shown, which is the industrial Internet of Things data classification device of this embodiment, including: An acquisition and processing module for real-time acquisition of sensor data and filtering processing; A feature analysis module for extracting time-frequency domain features of sensor data to obtain data window features; A type marking module for analyzing the comprehensive fluctuation index of the sensor data analysis window based on the sensor data, and marking the type of the sensor data according to the window comprehensive fluctuation index; A similarity analysis module for constructing a window feature vector based on the data window features, and analyzing the type similarity based on the window feature vector and the type marking of the sensor data; A data classification module for constructing a data classification model based on the data window features and the type marking of the sensor data to obtain data classification labels; A model update module for triggering the data classification model update rule based on the type similarity and the type marking of the sensor data.

[0074] This application embodiment also provides a computer-readable storage medium storing instructions, which when running on a computer, cause the computer to execute the industrial Internet of Things data classification method as described in the above method embodiment.

[0075] Those of ordinary skill in the art will understand that all or some of the steps and systems disclosed in the above methods can be implemented as software, firmware, hardware, and their appropriate combinations. Some physical components or all physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or can be implemented as hardware, or can be implemented as an integrated circuit, such as an application specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable programs, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that a communication medium typically contains computer-readable programs, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0076] Obviously, the above embodiments of the present invention are merely examples for clearly illustrating the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, based on the above description, other different forms of changes or variations can be made. It is impossible to enumerate all the implementation manners here. Any obvious changes or variations derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for classifying industrial Internet of Things data, characterized in that: include: Collect sensor data in real time and perform filtering; Extract the time-frequency domain features of the sensor data to obtain the data window features; Analyze the window comprehensive fluctuation index based on the sensor data, and mark the sensor data by type according to the window comprehensive fluctuation index; Building a window feature vector based on the data window features, and analyzing type similarity based on the window feature vector and the type tags of the sensor data; A data classification model is constructed based on the data window features and the type labels of the sensor data to obtain data classification labels; The data classification model update rules are triggered based on the type similarity and type labeling of sensor data.

2. The industrial Internet of Things data classification method according to claim 1 is characterized in that: According to the X-axis vibration speed in the three-axis vibration sensor data, the first window effective value is analyzed, and the first window effective value is compared with the speed threshold. If the first window effective value is greater than or equal to the speed threshold, the spectrum refinement analysis is triggered; otherwise, the spectrum refinement analysis is not triggered; the spectrum refinement analysis is to perform fast Fourier transform on the three-axis vibration sensor data to extract the rotation frequency and harmonic components of the monitored target, and analyze the first data window characteristics based on the rotation frequency and harmonic wave. The expression of the first data window characteristics is: Rf1=A2f+A3f / Af, where Rf1 represents the first data window characteristics, A2f represents the harmonic amplitude of 2 times the rotation frequency, A3f represents the harmonic amplitude of 3 times the rotation frequency, and Af represents the rotation frequency.

3. The industrial Internet of Things data classification method according to claim 2 is characterized in that: Analyze the second data window feature according to the infrared temperature sensor data, and take the ratio of the difference between the end temperature and the start temperature in the window and the window length as the second data window feature; The acoustic emission energy of each data point of the acoustic emission sensor data is compared with the energy threshold. If the acoustic emission energy is greater than or equal to the energy threshold, the current analysis data point is marked as an impact event; otherwise, the current analysis data point is not marked; According to the signal phase of the frequency conversion analysis, the expression of the signal phase is θ=2×π×Af, where θ represents the signal phase. The shock events whose signal phases of the data points corresponding to the shock events belong to [0,π / 2] are extracted as valid shock events, and the ratio of the number of valid shock events to the number of shock events is taken as the third data window feature.

4. The industrial Internet of Things data classification method according to claim 3 is characterized in that: The second window effective value is analyzed according to the Y-axis vibration velocity in the three-axis vibration sensor data, and the third window effective value is analyzed according to the Z-axis vibration velocity in the three-axis vibration sensor data. Dynamic weights are assigned to the first window effective value, the second window effective value, the third window effective value and the impact event within M minutes to obtain the window comprehensive fluctuation index. The window comprehensive fluctuation index is compared with the index threshold to mark the sensor data as fluctuation data and steady-state data.

5. The industrial Internet of Things data classification method according to claim 4 is characterized in that: A window feature vector is constructed based on the data window feature, and the cosine similarity between the window feature vector of the sensor data marked as fluctuating data and the window feature vector of the sensor data marked as steady-state data is used as the type similarity.

6. The industrial Internet of Things data classification method according to claim 5 is characterized in that: Analyze the characteristic abnormal state according to the second data window characteristic and the third data window characteristic. If the second data window characteristic is greater than or equal to the second characteristic threshold, set the characteristic abnormal state to temperature mutation abnormality; otherwise, set the characteristic abnormal state to normal temperature; if the third data window characteristic is greater than or equal to the third characteristic threshold, set the characteristic abnormal state to acoustic emission damage abnormality; otherwise, set the characteristic abnormal state to normal acoustic emission; A primary data classification label is constructed according to the data window feature. When the first data window feature is greater than the first feature threshold, the primary data classification label is set to gear abnormality and the secondary classification is triggered; when VZ(i) is greater than or equal to the kurtosis threshold, the primary data classification label is set to bearing abnormality and the secondary classification is triggered; when NAf / N is greater than or equal to the balance threshold, the primary data classification label is set to operation imbalance and the secondary classification is triggered; wherein NAf represents the number of data points with 1 times the rotation frequency, N represents the number of data points in the window, and VZ(i) represents the vibration speed of the Z axis; the secondary classification is: if at least two of the gear abnormality, bearing abnormality and operation imbalance exist in the primary data classification label at the same time, the secondary data classification label is set to emergency alarm; if the characteristic abnormal state is temperature mutation abnormality and acoustic emission damage abnormality, the secondary data classification label is set to high risk warning; if the characteristic abnormal state is temperature mutation abnormality or acoustic emission damage abnormality, the secondary data classification label is set to low risk warning; if the characteristic abnormal state is normal temperature and normal acoustic emission, the secondary data classification label is set to normal.

7. The industrial Internet of Things data classification method according to claim 6 is characterized in that: Extract sensor data within 1 hour. If the data points of the same sensor data are marked as fluctuation data and steady-state data in the window comprehensive fluctuation index analysis of different M-minute windows, the current analysis data point is used as the drift data point, and the ratio of the number of drift data points to the number of data points within 1 hour is used as the similarity drift compensation parameter. If the similarity drift compensation parameter is greater than or equal to the compensation threshold, the analysis process of type similarity is updated.

8. The industrial Internet of Things data classification method according to claim 7, characterized in that: The type similarity is compared with the similarity threshold. If the type similarity is less than or equal to the similarity threshold, the data classification model update rule is triggered; otherwise, the data classification model update rule is not triggered. The triggering data classification model update rule is to reduce the first feature threshold, the second feature threshold and the third feature threshold by G, where G represents the reduction ratio.

9. An industrial Internet of Things data classification device, characterized in that: include: The acquisition and processing module is used to collect sensor data in real time and perform filtering processing; The feature analysis module is used to extract the time-frequency domain features of the sensor data to obtain the data window features; A type marking module is used to analyze the window comprehensive fluctuation index based on the sensor data and mark the sensor data according to the window comprehensive fluctuation index; A similarity analysis module is used to construct a window feature vector based on the data window features, and analyze type similarity based on the window feature vector and the type label of the sensor data; A data classification module is used to build a data classification model based on the data window features and the type label of the sensor data to obtain a data classification label; The model updating module is used to trigger the data classification model updating rules according to the type similarity and the type label of the sensor data.

10. A storage medium, characterized in that: Instructions are stored, and when the instructions are run on a computer, the computer executes the industrial Internet of Things data classification method as described in any one of claims 1 to 8.

Images