A belt conveyor protection early warning method and system based on artificial intelligence

Abstract

The present invention discloses a belt conveyor protection and early warning method and system based on artificial intelligence, including constructing a belt conveyor sensor network to collect belt conveyor operation data and state data, processing the operation data to obtain main modal features and auxiliary modal features, matching common features, performing feature correction based on the matching results to obtain a first operation feature, performing a preliminary screening on the first operation feature, determining a first early warning strategy based on the preliminary screening results, inputting the state data into a state function to obtain a state factor, constructing a belt conveyor deep early warning model based on the preliminary screening results and the state factor, inputting the operation data and state data of the belt conveyor to be warned into the belt conveyor deep early warning model to obtain a hidden danger constant, performing a deep early warning based on the hidden danger constant to determine a second early warning strategy. This method can not only improve the accuracy of the belt conveyor protection and early warning method, but also has good interpretability, and can be directly applied to the belt conveyor protection and early warning system.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and early warning, and in particular to an artificial intelligence-based belt conveyor protection early warning method and system. Background Art

[0002] As an important material conveying equipment, belt conveyors are widely used in industrial production industries. Their efficient and stable operation plays a vital role in ensuring production continuity and improving production efficiency. Once a failure occurs, it will not only lead to production stagnation and economic losses, but may also cause safety accidents. Therefore, in the era of industrial automation and intelligence, various industries have put forward higher requirements for the protection and early warning technology of belt conveyors.

[0003] However, the current traditional belt conveyor protection and early warning technology has many shortcomings: first, traditional monitoring methods often rely on simple sensors and threshold judgments, which makes it difficult to conduct a comprehensive and accurate assessment of its complex operating status; at the same time, traditional technologies cannot detect and warn some potential fault hazards in advance; finally, traditional methods lack effective prediction and management of equipment maintenance and repair, resulting in high equipment maintenance costs and long downtime. The present invention uses artificial intelligence technology, combined with multimodal data fusion and deep learning algorithms, to conduct real-time analysis and deep mining of massive data during the operation of the belt conveyor, accurately identify various operating states of the belt conveyor, and conduct real-time monitoring and intelligent analysis, predict the possibility of failure in advance, and design a belt conveyor protection and early warning method and system based on artificial intelligence, which overcomes the shortcomings of traditional technology, improves the accuracy and reliability of belt conveyor protection and early warning, and can effectively reduce maintenance costs, ensure the safety and stable operation of production, and is of great significance to promoting the intelligent development of industrial production. Summary of the invention

[0004] The purpose of the present invention is to provide a belt conveyor protection early warning method and system based on artificial intelligence.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Constructing a belt conveyor sensor network to collect belt conveyor operation data and status data; the status data includes environmental status data and machine status data;

[0008] Processing the operation data to obtain a primary modal feature and an auxiliary modal feature, matching the common feature, and performing feature correction based on the matching result to obtain a first operation feature;

[0009] Performing a preliminary screening on the first operation feature, and determining a first early warning strategy according to the preliminary screening result; the first early warning strategy includes performing a deep early warning and outputting an operation safety result;

[0010] Input the state data into a state function to obtain a state factor, and construct a belt conveyor depth warning model according to the initial screening result and the state factor;

[0011] The operation data and status data of the belt conveyor to be warned are input into the belt conveyor deep warning model to obtain the hidden danger constant, and a deep warning is performed according to the hidden danger constant to determine the second warning strategy; the hidden danger constant includes a first hidden danger constant, a second hidden danger constant and a third hidden danger constant.

[0012] Furthermore, a method for constructing the belt conveyor sensor network to collect belt conveyor operation data and status data includes:

[0013] A visual sensing module, a laser scanning module, a thermal imaging module, a sound module and an environmental perception module are set up to form a belt conveyor sensor network;

[0014] The visual sensing module is composed of high-definition cameras at various positions of the belt conveyor, which is used to obtain image information of the belt conveyor in operation; the laser scanning module is composed of laser scanning sensors, which is used to obtain the three-dimensional point cloud data, contour information and laser beam reflection time of the belt conveyor; the thermal imaging module is composed of infrared thermal imaging sensors, which is used to detect the infrared radiation of the belt conveyor and convert it into temperature data of each component of the belt conveyor; the sound module is composed of sound sensors, which is used to obtain the sound wave signal generated during the operation of the belt conveyor; the sound wave signal includes working sound and noise;

[0015] The environmental perception module consists of a laser dust sensor, a temperature and humidity sensor, a micro-pressure difference sensor, and a vibration sensor, and is used to obtain the environmental status data of the belt conveyor;

[0016] The operation data of the belt conveyor is composed of the data collected by the visual sensing module, laser scanning module, thermal imaging module and sound module. The machine status data is obtained by directly extracting the belt conveyor status parameters and operation records, and the status data of the belt conveyor is composed of the machine status data and the environmental status data collected by the environmental perception module.

[0017] Furthermore, the method of processing the operating data to obtain the main modal features and the auxiliary modal features includes:

[0018] An image processing model is used to process the image information of the belt conveyor operation to obtain the main modal features; the image processing model includes a correction module, a belt appearance module, a belt deviation module, a color detection module, a product module, a safety module and an integration module;

[0019] The correction module performs denoising, enhancement, pixel correction and color correction operations on the image information of the belt conveyor to obtain a corrected image; the belt appearance module uses an edge detection algorithm to process the corrected image to obtain belt wear information, and uses threshold segmentation and morphological operations to process the corrected image to obtain belt crack information; the belt deviation module uses the RANSAC algorithm to fit the standard edge of the belt, and calculates the standard edge of the belt and the actual edge to obtain the belt deviation data; the color detection module performs color space conversion on the corrected image, and determines the color abnormality information of the equipment or product according to the color threshold; the product module calculates the product stacking situation based on the image depth information, uses the edge detection algorithm and the gray level co-occurrence matrix to extract the product shape features and texture features, and matches the product database according to the product shape features, texture features and color features to obtain product information; the safety module uses a deep learning human body detection model to detect unsafe factors of personnel, and uses a target detection model to detect unsafe factors of equipment and foreign objects around the belt conveyor; the integration module is connected with the correction module, belt appearance module, belt deviation module, color detection module, product module, and safety module, and integrates the information processed by the above modules to output the main modal features of the belt conveyor;

[0020] The collected data of the laser scanning module, the thermal imaging module and the sound module are processed to obtain auxiliary modal features. The specific method is as follows: the distance data is determined according to the time of laser beam reflection, the belt offset auxiliary data is determined by the distance data, the contour information and the three-dimensional point cloud data are integrated to obtain the current belt conveyor model, and the current belt conveyor model is compared with the standard belt model to obtain the belt wear and deformation auxiliary data; the temperature distribution diagram is drawn according to the temperature data of each component of the belt conveyor, and the thermal change trend is calculated by a statistical method; the vibration signal of the vibration sensor is used to correct the sound wave signal, and the corrected sound wave signal is processed by Fourier transform to identify the working sound and noise, and feature extraction is performed to obtain the operating sound wave feature and the noise feature, and the comprehensive similarity of the combined vector of the operating sound wave feature and the noise feature and the historical sound wave vector of the sound wave database is calculated to obtain the belt conveyor part feature; the sound wave database includes historical sound wave vectors and corresponding part features.

[0021] Furthermore, the method for obtaining the first operating characteristic includes:

[0022] Calculate the comprehensive similarity between the auxiliary modal feature vectors, match the auxiliary modal features with the auxiliary modal features according to the highest comprehensive similarity, fuse the two auxiliary modal features according to the matching result to obtain the auxiliary modal correction feature, and combine the auxiliary modal correction feature and the unmatched auxiliary modal feature to form the first auxiliary modal feature;

[0023] Calculate the comprehensive similarity between the first auxiliary modal feature and the main modal feature, determine the second auxiliary modal feature by the highest comprehensive similarity, calculate the correlation between the main modal feature and the unmatched first auxiliary modal feature, take the unmatched first auxiliary modal feature greater than the correlation threshold as the third auxiliary modal feature, compare the second auxiliary modal feature with the main modal feature to obtain the degree of difference, cross-check the third auxiliary modal feature with the main modal feature to obtain the first main modal feature, perform feature fusion based on the first main modal feature, the second auxiliary modal feature and the degree of difference to obtain the second main modal feature, and form the second main modal feature, the unmatched first auxiliary modal feature and the unmatched main modal feature into a first operating feature; the second auxiliary modal feature and the main modal feature are the same index collected and calculated by different sensing modules; the third auxiliary modal feature is a different index that affects the main modal feature collected and calculated by different sensing modules;

[0024] The mean, standard deviation and skewness corresponding to the temperature information, acoustic wave characteristics and frequency band offset in the first operating characteristic are statistically analyzed, and the temperature stability, acoustic wave stability and offset stability are calculated based on the statistical results. The comprehensive stability is obtained by weighted fusion of various types of stabilities.

[0025] Furthermore, the method for determining the first early warning strategy includes:

[0026] Hierarchical clustering is used to divide the first operation characteristics into belt damage hidden danger characteristics, accident quality hidden danger characteristics and potential hidden danger characteristics, and the multi-sphere method is used to preliminarily screen the first operation characteristics, and the first early warning strategy is determined according to the preliminary screening results; the number of spheres in the multi-sphere method is determined by the hierarchical clustering results; the preliminary screening results include abnormal operation characteristics and normal operation characteristics;

[0027] The first early warning strategy is specifically:

[0028] A first-level warning is directly issued based on abnormal operating characteristics and corresponding feature labels; the ratio of normal characteristics to corresponding safety thresholds is calculated to obtain the operating characteristic risk, and the mean values ​​of operating characteristic risks of different categories are calculated according to the first operating characteristic classification results. When the mean operating characteristic risk is greater than 60%, a deep warning is issued to determine the second warning strategy, otherwise, the operating safety results are output; the first-level warning indicates that the first operating characteristic exceeds the safety threshold, and safety measures must be taken immediately according to the specific characteristics.

[0029] Furthermore, the method of inputting the state data into the state function to obtain the state factor includes:

[0030] The state factors include environmental state factors and machine state factors;

[0031] Input the environmental state data into the environmental state function to obtain the environmental state factor, the expression is:

[0032]

[0033] in is the environmental state factor, is the dust accumulation weight, is the temperature and humidity weight, To measure the impact weight, is the dust concentration, is the current air velocity, is the standard air flow rate, is the current wind pressure, is the standard wind pressure, is the current ambient temperature, is the standard ambient temperature, is the standard deviation of ambient temperature, Current ambient humidity, is the standard ambient humidity, is the standard deviation of ambient humidity, is the indoor light intensity, is the indoor magnetic field strength, is the mean value of mechanical vibration intensity, is the standard mechanical vibration intensity;

[0034] Input the machine state data into the machine state function to obtain the machine state factor, which is expressed as:

[0035]

[0036] in is the machine state factor, For power supply influence weight, is the work efficiency weight, Maintain weights for operations, is the average current of the device, is the maximum rated current of the device, is the current standard deviation, is the maximum rated voltage of the device, is the mean value of the device voltage, is the voltage standard deviation, For current work efficiency, For maximum work efficiency, is the current material flow rate, is the base material flow rate, is the current continuous working time of the device, Rated continuous working time of the equipment, is the strip replacement time, Set a usage time for the strip, is the number of emergency stop operations of the equipment, Manual frequency tuning times for the device. The lubrication interval for the strips is Specify lubrication times for the strips.

[0037] Furthermore, the method for obtaining the hidden danger constant includes:

[0038] The belt damage hidden danger characteristics and belt tension form a belt quality feature set; the accident quality hidden danger characteristics and machine status factors form an accident quality feature set; the potential hidden danger characteristics, environmental status factors and comprehensive stability form a comprehensive quality feature set;

[0039] Constructing a belt conveyor depth warning model, the belt conveyor depth warning model includes an input layer, a base model layer and an output layer; the base model layer includes a linear regression base model, a BP neural network base model and a multimodal decision tree base model;

[0040] The linear regression base model performs performance linear regression on the belt damage hidden danger characteristics and belt tension to predict the first hidden danger constant, uses polynomial feature expansion to process crack information, uses Huber loss function for robustness optimization, uses Adam optimizer to adjust the learning rate, uses belt quality feature set to train the model and verify the model performance;

[0041] The BP neural network based model performs nonlinear regression on accident quality hidden danger characteristics and machine status factors to predict the second hidden danger constant. The LSTM+Attention mechanism is used to capture the time dependence of accident quality hidden danger characteristics. The cross entropy loss function is used to measure the difference between the probability distribution of the model output and the true distribution. Dynamic regularization is used to automatically adjust the Dropout rate according to the training error. The accident quality feature set is used to train the model and verify the model performance.

[0042] The multimodal decision tree-based model adopts a multi-head cross-attention mechanism to fuse potential hidden danger features to obtain potential fusion hidden danger features, adopts a decision tree to capture the complex patterns and interactions of potential fusion hidden danger features, environmental state factors and comprehensive stability, performs nonlinear regression to predict the third hidden danger constant, adopts dynamic pruning to adjust the tree depth, adopts weighted classification loss and regression loss to measure model performance, adopts RMSprop optimizer to adjust the learning rate, adopts a comprehensive quality feature set to train the model and verify the model performance; the splitting criterion of the decision tree is specifically as follows:

[0043]

[0044] in For the splitting criterion, is the Gini coefficient, is the feature weight, is the interaction weight, is the state weight, is the feature importance, , is the interactive feature among the potential fusion risk features, is the machine state factor;

[0045] The output layer is connected to the three base models, receiving the prediction results of the base models and outputting the hidden danger constants.

[0046] Furthermore, the method for performing a deep warning according to the hidden danger constant to determine a second warning strategy includes:

[0047] The second warning strategy includes performing a second-level warning, performing a third-level warning, and outputting operation safety; the second-level warning and the third-level warning indicate that the first operation characteristics of the same type cumulatively interact with each other and there is a potential risk;

[0048] When the hidden danger constant is greater than the corresponding hidden danger constant threshold, a second-level warning is issued for the same category of operation;

[0049] When there is only one hidden danger constant less than or equal to the corresponding hidden danger constant threshold, the ratio of the same type of hidden danger constant to the corresponding hidden danger constant threshold is calculated to obtain the hidden danger constant ratio. If the hidden danger constant ratio is greater than the first ratio threshold, a third-level warning is issued, otherwise the corresponding type of operation is output as safe.

[0050] When the remaining two hidden danger constants are less than or equal to the corresponding hidden danger constant threshold, the other hidden danger constant is input into the weight function to obtain the hidden danger constant weight, and the hidden danger constant weight and the product of the two hidden danger constants are calculated to obtain the weighted hidden danger constant. If the weighted hidden danger constant is greater than the corresponding hidden danger constant threshold, a third-level warning is issued, otherwise the corresponding class operation safety is output;

[0051] When the three hidden danger constants are all less than or equal to the corresponding hidden danger constant thresholds, the ratio of each type of hidden danger constant is calculated. If the mean of the hidden danger constant ratio is greater than the second ratio threshold, a third-level warning is issued. Otherwise, the overall operation of the output belt conveyor is safe.

[0052] In the second aspect, a belt conveyor protection early warning system based on artificial intelligence includes:

[0053] Data acquisition module: used to build a belt conveyor sensor network to collect belt conveyor operation data and status data;

[0054] Data processing module: used for processing the operation data to obtain the main modal features and the auxiliary modal features, matching the common features, performing feature correction based on the matching results to obtain the first operation features, and inputting the state data into the state function to obtain the state factor;

[0055] A primary warning module: used for preliminarily screening the first operation characteristics and determining a first warning strategy according to the preliminarily screening results;

[0056] Deep warning module: used to construct a belt conveyor deep warning model according to the initial screening results and the state factors, input the operation data and state data of the belt conveyor to be warned into the belt conveyor deep warning model to obtain the hidden danger constant, and perform deep warning according to the hidden danger constant to determine the second warning strategy;

[0057] Management module: used to manage the operation data, the status data, the preliminary screening results and the hidden danger constants, and to make a first-level warning according to the preliminary screening results and a second-level warning according to the hidden danger constants.

[0058] The beneficial effects of the present invention are:

[0059] The present invention is a belt conveyor protection early warning method and system based on artificial intelligence. Compared with the prior art, the present invention has the following technical effects:

[0060] The present invention can improve the accuracy of belt conveyor protection warning by constructing a belt conveyor sensor network, feature matching and correction, initial screening and warning, constructing state factors, building models and in-depth analysis of warning steps, thereby improving the accuracy of belt conveyor protection warning, and optimizing belt conveyor protection warning, which can greatly save resources and improve work efficiency. It can realize belt conveyor protection warning and perform real-time data monitoring and protection warning on belt conveyors, which is of great significance to improving the efficiency of belt conveyor operation data monitoring and reducing belt conveyor safety hazards and maintenance costs. It can adapt to different belt conveyor protection warning systems and belt conveyor protection warning needs of different users, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 The present invention is a flowchart of the steps of a belt conveyor protection early warning method based on artificial intelligence. DETAILED DESCRIPTION

[0062] The present invention is further described below by means of specific embodiments. The illustrative embodiments and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0063] The present invention provides a belt conveyor protection early warning method and system based on artificial intelligence, comprising the following steps:

[0064] like Figure 1 As shown, in this embodiment, the following steps are included:

[0065] Constructing a belt conveyor sensor network to collect belt conveyor operation data and status data; the status data includes environmental status data and machine status data;

[0066] Processing the operation data to obtain a primary modal feature and an auxiliary modal feature, matching the common feature, and performing feature correction based on the matching result to obtain a first operation feature;

[0067] Performing a preliminary screening on the first operation feature, and determining a first early warning strategy according to the preliminary screening result; the first early warning strategy includes performing a deep early warning and outputting an operation safety result;

[0068] Input the state data into a state function to obtain a state factor, and construct a belt conveyor depth warning model according to the initial screening result and the state factor;

[0069] The operation data and status data of the belt conveyor to be warned are input into the belt conveyor deep warning model to obtain the hidden danger constant, and a deep warning is performed according to the hidden danger constant to determine the second warning strategy; the hidden danger constant includes a first hidden danger constant, a second hidden danger constant and a third hidden danger constant.

[0070] In this embodiment, the method for constructing the belt conveyor sensor network to collect belt conveyor operation data and status data includes:

[0071] A visual sensing module, a laser scanning module, a thermal imaging module, a sound module and an environmental perception module are set up to form a belt conveyor sensor network;

[0072] The visual sensing module is composed of high-definition cameras at various positions of the belt conveyor, which is used to obtain image information of the belt conveyor in operation; the laser scanning module is composed of laser scanning sensors, which is used to obtain the three-dimensional point cloud data, contour information and laser beam reflection time of the belt conveyor; the thermal imaging module is composed of infrared thermal imaging sensors, which is used to detect the infrared radiation of the belt conveyor and convert it into temperature data of each component of the belt conveyor; the sound module is composed of sound sensors, which is used to obtain the sound wave signal generated during the operation of the belt conveyor; the sound wave signal includes working sound and noise;

[0073] The environmental perception module consists of a laser dust sensor, a temperature and humidity sensor, a micro-pressure difference sensor, and a vibration sensor, and is used to obtain the environmental status data of the belt conveyor;

[0074] The operation data of the belt conveyor is composed of the data collected by the visual sensing module, laser scanning module, thermal imaging module and sound module; the machine status data is obtained by directly extracting the state parameters and operation records of the belt conveyor, and the state data of the belt conveyor is composed of the machine status data and the environmental status data collected by the environmental perception module;

[0075] In actual evaluation, the environmental status data include mechanical vibration, dust concentration in the surrounding environment, wind conditions, light intensity, magnetic field intensity, mechanical vibration intensity, and temperature and humidity; the machine status data include equipment current and voltage, equipment efficiency, material flow, continuous working time, and operation and maintenance parameters;

[0076] Take a belt conveyor that transports products in a factory as an example to obtain operation data and status data.

[0077] In this embodiment, the method of processing the operating data to obtain the main modal features and the auxiliary modal features includes:

[0078] An image processing model is used to process the image information of the belt conveyor operation to obtain the main modal features; the image processing model includes a correction module, a belt appearance module, a belt deviation module, a color detection module, a product module, a safety module and an integration module;

[0079] The correction module performs denoising, enhancement, pixel correction and color correction operations on the image information of the belt conveyor to obtain a corrected image; the belt appearance module uses an edge detection algorithm to process the corrected image to obtain belt wear information, and uses threshold segmentation and morphological operations to process the corrected image to obtain belt crack information; the belt deviation module uses the RANSAC algorithm to fit the standard edge of the belt, and calculates the standard edge of the belt and the actual edge to obtain the belt deviation data; the color detection module performs color space conversion on the corrected image, and determines the color abnormality information of the equipment or product according to the color threshold; the product module calculates the product stacking situation based on the image depth information, uses the edge detection algorithm and the gray level co-occurrence matrix to extract the product shape features and texture features, and matches the product database according to the product shape features, texture features and color features to obtain product information; the safety module uses a deep learning human body detection model to detect unsafe factors of personnel, and uses a target detection model to detect unsafe factors of equipment and foreign objects around the belt conveyor; the integration module is connected with the correction module, belt appearance module, belt deviation module, color detection module, product module, and safety module, and integrates the information processed by the above modules to output the main modal features of the belt conveyor;

[0080] Processing the collected data of the laser scanning module, the thermal imaging module and the sound module to obtain auxiliary modal features, the specific method is: determining the distance data according to the time of laser beam reflection, determining the belt offset auxiliary data from the distance data, fusing the contour information and the three-dimensional point cloud data to obtain the current belt conveyor model, and comparing the current belt conveyor model with the standard belt model to obtain the belt wear and deformation auxiliary data; drawing a temperature distribution diagram according to the temperature data of each component of the belt conveyor, and using statistical methods to calculate the thermal change trend; using the vibration signal of the vibration sensor to correct the sound wave signal, using Fourier transform to process the corrected sound wave signal to identify the working sound and noise, performing feature extraction to obtain the operating sound wave feature and noise feature, and calculating the comprehensive similarity between the combined vector of the operating sound wave feature and the noise feature and the historical sound wave vector of the sound wave database to obtain the belt conveyor part feature; the sound wave database includes historical sound wave vectors and corresponding part features;

[0081] In the actual evaluation, the main modal features include belt wear information, belt crack information, belt deviation data, color abnormality information, product accumulation, product information, unsafe factors and foreign matter conditions; the unsafe factors include personnel unsafe factors and equipment unsafe factors; the auxiliary modal features include belt deviation auxiliary data, belt wear and deformation auxiliary data, temperature distribution diagram, thermal change trend and belt conveyor part characteristics;

[0082] Processing the operating data to obtain primary modal features and auxiliary modal features:

[0083] Visual sensor module (consisting of 5 high-definition cameras): belt wear depth [0.1, 0.15, 0.2] / mm, belt wear area [5, 8, 6] / cm², belt crack numbers [5, 2, 1], crack information [(3h, 4h, 8d, 6d, 2v), (5h, 4v, 3v), (4d, 2v,)], (h represents horizontal cracks, v represents vertical cracks, d represents oblique cracks, and the number represents length in cm), belt offset data is [5, 3, 4] / cm, color anomaly information is [(10, 20, 30), (0, 0, 0 ),(15,25,35)] (indicates the offset from the standard RGB value. The higher the temperature, the greater the color offset). The product stacking is [(30,20,15),(25,18,12),(35,22,18)] / cm (indicates the length, width and height of the stacking). The product information is [A,B,A]. The distance between the personnel and the conveyor belt is [0.5,2,1] / m respectively. The personnel safety equipment is [no helmet, fully worn, no gloves]. The surrounding risk equipment is [none, high temperature equipment, no], and the foreign matter is [none, metal block, no].

[0084] Laser scanning module: belt deviation auxiliary data is [4, 6, 5] / cm, belt wear deformation auxiliary data is [0.05, 0.1, 0.08] / mm;

[0085] Thermal imaging module: The temperature of the belt conveyor head is 45°C, the middle is 50°C, and the tail is 48°C. The thermal change trend calculated by statistical methods is [0.5°C / h, -0.3°C / h, 0.2°C / h];

[0086] Sound module: The running sound wave characteristics are [0.8, 0.7, 0.9], the noise characteristics are [0.3, 0.2, 0.4], and the belt conveyor part characteristics obtained after comprehensive similarity calculation with the historical sound wave vectors in the sound wave database are [normal head, abnormal middle, normal tail];

[0087] Environmental perception module: dust concentration 0.05mg / m³, current air flow rate 2m / s, standard air flow rate 1.5m / s, current wind pressure 100Pa, standard wind pressure 90Pa, current ambient temperature 30°C, standard ambient temperature 25°C, ambient temperature standard deviation 2°C, current ambient humidity 60%, standard ambient humidity 50%, ambient humidity standard deviation 5%, indoor light intensity 800lux, indoor magnetic field intensity 0.02T, mechanical vibration intensity mean 0.03m / s², standard mechanical vibration intensity 0.02m / s²;

[0088] The machine status data are: equipment current average 10A, equipment maximum rated current 15A, current standard deviation 1A, equipment maximum rated voltage 380V, equipment voltage average 370V, voltage standard deviation 5V, current working efficiency 0.8, maximum working efficiency 1, current material flow 50t / h, benchmark material flow 60t / h, equipment current continuous working time 10h, equipment rated continuous working time 12h, strip replacement time 500h, strip specified use time 600h, equipment emergency stop operation times 2 times, equipment manual frequency adjustment times 3 times, strip lubrication time interval 100h, strip specified lubrication time 80h.

[0089] In this embodiment, the method for obtaining the first operating characteristic includes:

[0090] Calculate the comprehensive similarity between the auxiliary modal feature vectors, match the auxiliary modal features with the auxiliary modal features according to the highest comprehensive similarity, fuse the two auxiliary modal features according to the matching result to obtain the auxiliary modal correction feature, and combine the auxiliary modal correction feature and the unmatched auxiliary modal feature to form the first auxiliary modal feature;

[0091] Calculate the comprehensive similarity between the first auxiliary modal feature and the main modal feature, determine the second auxiliary modal feature by the highest comprehensive similarity, calculate the correlation between the main modal feature and the unmatched first auxiliary modal feature, take the unmatched first auxiliary modal feature greater than the correlation threshold as the third auxiliary modal feature, compare the second auxiliary modal feature with the main modal feature to obtain the degree of difference, cross-check the third auxiliary modal feature with the main modal feature to obtain the first main modal feature, perform feature fusion based on the first main modal feature, the second auxiliary modal feature and the degree of difference to obtain the second main modal feature, and form the second main modal feature, the unmatched first auxiliary modal feature and the unmatched main modal feature into a first operating feature; the second auxiliary modal feature and the main modal feature are the same index collected and calculated by different sensing modules; the third auxiliary modal feature is a different index that affects the main modal feature collected and calculated by different sensing modules;

[0092] The mean, standard deviation and skewness corresponding to the temperature information, acoustic wave characteristics and belt offset in the first operation characteristic are counted, and the temperature stability, acoustic wave stability and offset stability are calculated according to the statistical results, and the comprehensive stability is obtained by weighted fusion of various stabilities;

[0093] In the actual evaluation, the improved cosine similarity (taking into account the physical meaning of the features and data distribution) calculated that the comprehensive similarity between the belt offset auxiliary data and the belt offset is 0.82, the comprehensive similarity between the belt wear and deformation auxiliary data and the belt wear information is 0.75, the comprehensive similarity between the temperature distribution map and the temperature data of each component of the belt conveyor is 0.68, and the comprehensive similarity between the combined vector of the operating sound wave feature and the noise feature and the historical sound wave vector of the sound wave database is 0.73. According to the highest comprehensive similarity of 0.82, the belt offset auxiliary data and the belt offset are matched for feature fusion to obtain the auxiliary modal correction feature. The fused feature is [4.4, 4.1, 4.7] / cm;

[0094] The comprehensive similarity between the first auxiliary modal feature and the main modal feature is calculated, among which the highest comprehensive similarity with the belt offset data is 0.8, and the second auxiliary modal feature is determined to be the belt offset data; the correlation between the main modal feature and the unmatched first auxiliary modal feature is calculated, and the correlation threshold is determined to be 0.65, among which the correlation between the belt wear depth information and the belt offset auxiliary data is 0.7>0.65, which is used as the third auxiliary modal feature; the second auxiliary modal feature (belt offset data) is compared with the main modal feature, and the difference degree is [0.6,-0.9,0.3] / cm (indicating the difference value); the third auxiliary modal feature (Belt wear depth information) is cross-referenced with the main modal features to obtain a new wear depth feature combination [0.13, 0.17, 0.21] / mm as the first main modal feature; based on the first main modal feature, the second auxiliary modal feature and the degree of difference, feature fusion is performed to obtain the second main modal feature. For example, the fused feature is [4.7, 3.4, 4.4, 0.15, 0.18, 0.2] / cm (including comprehensive information such as offset and wear), and the second main modal feature is combined with the unmatched first auxiliary modal feature and the unmatched main modal feature (such as color abnormality information, product accumulation, etc.) to form the first operating feature;

[0095] The mean, standard deviation and skewness of the temperature information, acoustic wave characteristics and belt offset in the first operating feature are (47.67, 0.8, 4.5), (2.08, 0.008, 0.45) and (0.2, 0.1, 0.15) respectively. The temperature stability is calculated to be 0.8, the acoustic wave stability is 0.9, and the offset stability is 0.85. Assuming the weights are 0.3, 0.3 and 0.4 respectively, the weighted fusion is performed to obtain a comprehensive stability of 0.845.

[0096] In this embodiment, the method for determining the first early warning strategy includes:

[0097] Hierarchical clustering is used to divide the first operation characteristics into belt damage hidden danger characteristics, accident quality hidden danger characteristics and potential hidden danger characteristics, and the multi-sphere method is used to preliminarily screen the first operation characteristics, and the first early warning strategy is determined according to the preliminary screening results; the number of spheres in the multi-sphere method is determined by the hierarchical clustering results; the preliminary screening results include abnormal operation characteristics and normal operation characteristics;

[0098] The first early warning strategy is specifically:

[0099] Directly issue a first-level warning based on abnormal operation characteristics and corresponding feature labels; calculate the ratio of normal characteristics to corresponding safety thresholds to obtain the operation characteristic risk, calculate the average of different types of operation characteristic risk according to the first operation characteristic classification results, and issue a deep warning to determine the second warning strategy when the average of the operation characteristic risk is greater than 60%, otherwise, output the operation safety result; the first-level warning indicates that the first operation characteristic exceeds the safety threshold, and safety measures must be taken immediately according to the specific characteristics;

[0100] In the actual assessment, after clustering, the belt damage hidden danger characteristics are [0.2mm, 3, 4.7cm] (including wear, crack quantity and offset), the accident quality hidden danger characteristics are [0.8, 50t / h, 0.8] (including efficiency and material flow and other characteristics), and the potential hidden danger characteristics are [0.845, 0.05mg / m³, 2m / s] (including comprehensive stability and environmental data and other characteristics). The multi-sphere method is used for preliminary screening. The wear depth, offset and crack quantity exceed the corresponding thresholds of 0.18mm, 4cm and 2, and these characteristics are directly issued a first-level warning;

[0101] The risk of the operating characteristics is obtained by calculating the ratio of the normal characteristics to the corresponding safety thresholds. For example, in the accident quality hazard characteristics, the work efficiency and the corresponding safety thresholds are 0.8 and 0.85, and the material flow and the corresponding thresholds are 50t / h and 55t / h. The work efficiency risk is 0.94, and the material flow risk is 0.91. The average risk of this type of operating characteristics is 0.925. Therefore, a deep warning is issued for this type of operating characteristics.

[0102] In this embodiment, the method of inputting the state data into the state function to obtain the state factor includes:

[0103] The state factors include environmental state factors and machine state factors;

[0104] Input the environmental state data into the environmental state function to obtain the environmental state factor, the expression is:

[0105]

[0106] in is the environmental state factor, is the dust accumulation weight, is the temperature and humidity weight, To measure the impact weight, is the dust concentration, is the current air velocity, is the standard air flow rate, is the current wind pressure, is the standard wind pressure, is the current ambient temperature, is the standard ambient temperature, is the standard deviation of ambient temperature, Current ambient humidity, is the standard ambient humidity, is the standard deviation of ambient humidity, is the indoor light intensity, is the indoor magnetic field strength, is the mean value of mechanical vibration intensity, is the standard mechanical vibration intensity;

[0107] Input the machine state data into the machine state function to obtain the machine state factor, which is expressed as:

[0108]

[0109] in is the machine state factor, For power supply influence weight, is the work efficiency weight, Maintain weights for operations, is the average current of the device, is the maximum rated current of the device, is the current standard deviation, is the maximum rated voltage of the device, is the mean value of the device voltage, is the voltage standard deviation, For current work efficiency, For maximum work efficiency, is the current material flow rate, is the base material flow rate, is the current continuous working time of the device, Rated continuous working time of the equipment, is the strip replacement time, Set a usage time for the strip, is the number of emergency stop operations of the equipment, Manual frequency tuning times for the device. For the strip lubrication interval, Specify lubrication times for the strips;

[0110] In the actual evaluation, the dust accumulation weight, temperature and humidity weight, and measurement influence weight are taken as 0.3, 0.4, and 0.3, and the collected environmental status data are substituted into the above formula to calculate the environmental status factor of 0.58; the power supply influence weight, work efficiency weight, and operation and maintenance weight are taken as 0.2, 0.4, and 0.4, and the collected machine status data are substituted into the above formula to calculate the machine status factor of 0.58.

[0111] In this embodiment, the method for obtaining the hidden danger constant includes:

[0112] The belt damage hidden danger characteristics and belt tension form a belt quality feature set; the accident quality hidden danger characteristics and machine status factors form an accident quality feature set; the potential hidden danger characteristics, environmental status factors and comprehensive stability form a comprehensive quality feature set;

[0113] Constructing a belt conveyor depth warning model, the belt conveyor depth warning model includes an input layer, a base model layer and an output layer; the base model layer includes a linear regression base model, a BP neural network base model and a multimodal decision tree base model;

[0114] The linear regression base model performs performance linear regression on the belt damage hidden danger characteristics and belt tension to predict the first hidden danger constant, uses polynomial feature expansion to process crack information, uses Huber loss function for robustness optimization, uses Adam optimizer to adjust the learning rate, uses belt quality feature set to train the model and verify the model performance;

[0115] The BP neural network based model performs nonlinear regression on accident quality hidden danger characteristics and machine status factors to predict the second hidden danger constant. The LSTM+Attention mechanism is used to capture the time dependence of accident quality hidden danger characteristics. The cross entropy loss function is used to measure the difference between the probability distribution of the model output and the true distribution. Dynamic regularization is used to automatically adjust the Dropout rate according to the training error. The accident quality feature set is used to train the model and verify the model performance.

[0116] The multimodal decision tree-based model adopts a multi-head cross-attention mechanism to fuse potential hidden danger features to obtain potential fusion hidden danger features, adopts a decision tree to capture the complex patterns and interactions of potential fusion hidden danger features, environmental state factors and comprehensive stability, performs nonlinear regression to predict the third hidden danger constant, adopts dynamic pruning to adjust the tree depth, adopts weighted classification loss and regression loss to measure model performance, adopts RMSprop optimizer to adjust the learning rate, adopts a comprehensive quality feature set to train the model and verify the model performance; the splitting criterion of the decision tree is specifically as follows:

[0117]

[0118] in For the splitting criterion, is the Gini coefficient, is the feature weight, is the interaction weight, is the state weight, is the feature importance, , is the interactive feature among the potential fusion risk features, is the machine state factor;

[0119] The output layer is connected to the three base models, receiving the prediction results of the base models and outputting the hidden danger constants;

[0120] In the actual evaluation, the belt quality feature set was randomly divided into the first training set and the first test set according to a ratio of 7:3. The first training set was used to train the model, and the first test set was used to verify the model performance. The accident quality feature set was randomly divided into the second training set and the second test set according to a ratio of 7:3. The second training set was used to train the model, and the second test set was used to verify the model performance. The comprehensive quality feature set was randomly divided into the third training set and the third test set according to a ratio of 7:3. The third training set was used to train the model, and the third test set was used to verify the model performance.

[0121] The crack information is processed by polynomial feature expansion. The number of cracks 2, transverse crack length 3cm, and longitudinal crack length 4cm are expanded to [2, 3, 4, 6, 12, 8] (considering the product and square of the crack length), and the wear depth 0.2mm is expanded to [0.2, 0.04, 0.08] (considering the square and cube of the wear depth). The potential hidden danger feature [0.845, 0.05, 2] is fused with the environmental state factor 0.58 and the comprehensive stability 0.845 to obtain the new feature [0.73, 0.045, 1.90.845]. The Gini coefficient, feature weight, interaction weight, and state weight are 0.4, 0.2, 0.3, and 0.1, respectively.

[0122] The first hidden danger constant, the second hidden danger constant and the third hidden danger constant output by the belt conveyor deep warning model are 0.45, 0.52 and 0.63 respectively.

[0123] In this embodiment, the method for determining the second warning strategy by performing a deep warning according to the hidden danger constant includes:

[0124] The second warning strategy includes performing a second-level warning, performing a third-level warning, and outputting operation safety; the second-level warning and the third-level warning indicate that the first operation characteristics of the same type cumulatively interact with each other and there is a potential risk;

[0125] When the hidden danger constant is greater than the corresponding hidden danger constant threshold, a second-level warning is issued for the same category of operation;

[0126] When there is only one hidden danger constant less than or equal to the corresponding hidden danger constant threshold, the ratio of the same type of hidden danger constant to the corresponding hidden danger constant threshold is calculated to obtain the hidden danger constant ratio. If the hidden danger constant ratio is greater than the first ratio threshold, a third-level warning is issued, otherwise the corresponding type of operation is output as safe.

[0127] When the remaining two hidden danger constants are less than or equal to the corresponding hidden danger constant threshold, the other hidden danger constant is input into the weight function to obtain the hidden danger constant weight, and the hidden danger constant weight and the product of the two hidden danger constants are calculated to obtain the weighted hidden danger constant. If the weighted hidden danger constant is greater than the corresponding hidden danger constant threshold, a third-level warning is issued, otherwise the corresponding class operation safety is output;

[0128] When the three hidden danger constants are all less than or equal to the corresponding hidden danger constant threshold, the ratio of each type of hidden danger constant is calculated. If the mean of the hidden danger constant ratio is greater than the second ratio threshold, a third-level warning is issued. Otherwise, the overall operation of the output belt conveyor is safe.

[0129] In actual assessment, when the first hidden danger constant is greater than the first hidden danger constant threshold, it indicates that the belt of the conveyor belt is at risk of damage or breakage, and it is necessary to adjust the appropriate belt tension and check the causes of belt scratches and wear; when the second hidden danger constant is greater than the second hidden danger constant threshold, it indicates that there is a risk of equipment damage and production and transportation accidents during the operation of the conveyor belt, and it is necessary to check the belt deviation, loose or damaged parts, and product accumulation on the conveyor belt; when the third hidden danger constant is greater than the third hidden danger constant threshold, it indicates that there are fire, personnel, environment or equipment unsafe factors, and it is necessary to adjust the belt conveyor operation parameters and check the factors involved in 5S management;

[0130] The corresponding hidden danger constant thresholds are 0.6, 0.6, and 0.6 respectively, so a second-level warning is given to the potential hidden danger characteristics as a whole. The calculated hidden danger constant weight is 1.916, so the first weighted hidden danger constant and the second weighted hidden danger constant are 0.864 and 0.998 respectively. The corresponding weighted hidden danger constant thresholds are 0.9 and 0.9, so the belt quality characteristics are output as "operation safety", and a third-level warning is given to the accident quality hidden danger characteristics as a whole.

[0131] In the second aspect, a belt conveyor protection early warning system based on artificial intelligence includes:

[0132] Data acquisition module: used to build a belt conveyor sensor network to collect belt conveyor operation data and status data;

[0133] Data processing module: used for processing the operation data to obtain the main modal features and the auxiliary modal features, matching the common features, performing feature correction based on the matching results to obtain the first operation features, and inputting the state data into the state function to obtain the state factor;

[0134] A primary warning module: used for preliminarily screening the first operation characteristics and determining a first warning strategy according to the preliminarily screening results;

[0135] Deep warning module: used to construct a belt conveyor deep warning model according to the initial screening results and the state factors, input the operation data and state data of the belt conveyor to be warned into the belt conveyor deep warning model to obtain the hidden danger constant, and perform deep warning according to the hidden danger constant to determine the second warning strategy;

[0136] Management module: used to manage the operation data, the status data, the preliminary screening results and the hidden danger constants, and to make a first-level warning according to the preliminary screening results and a second-level warning according to the hidden danger constants.

[0137] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A belt conveyor protection early warning method based on artificial intelligence, characterized in that: The following steps are involved: S1. Constructing a belt conveyor sensor network to collect belt conveyor operation data and status data; the status data includes environmental status data and machine status data; S2. Processing the operation data to obtain main modal features and auxiliary modal features, matching common features, and performing feature correction based on the matching results to obtain a first operation feature; S3, performing a preliminary screening on the first operation characteristic, and determining a first early warning strategy according to the preliminary screening result; The first early warning strategy includes conducting a deep early warning and outputting an operation safety result; S4, inputting the state data into a state function to obtain a state factor, and constructing a belt conveyor depth warning model according to the initial screening result and the state factor; S5, inputting the operation data and status data of the belt conveyor to be warned into the belt conveyor deep warning model to obtain the hidden danger constant, and performing deep warning according to the hidden danger constant to determine the second warning strategy; The hidden danger constants include a first hidden danger constant, a second hidden danger constant and a third hidden danger constant; The method for determining the first early warning strategy includes: Hierarchical clustering is used to divide the first operation characteristics into belt damage hidden danger characteristics, accident quality hidden danger characteristics and potential hidden danger characteristics, and the multi-sphere method is used to preliminarily screen the first operation characteristics, and the first early warning strategy is determined according to the preliminary screening results; the number of spheres in the multi-sphere method is determined by the hierarchical clustering results; the preliminary screening results include abnormal operation characteristics and normal operation characteristics; The first early warning strategy is specifically: A first-level warning is directly issued based on abnormal operating characteristics and corresponding feature labels; the ratio of normal characteristics to corresponding safety thresholds is calculated to obtain the operating characteristic risk, and the mean values ​​of operating characteristic risks of different categories are calculated according to the first operating characteristic classification results. When the mean operating characteristic risk is greater than 60%, a deep warning is issued to determine the second warning strategy, otherwise, the operating safety results are output; the first-level warning indicates that the first operating characteristic exceeds the safety threshold, and safety measures must be taken immediately according to the specific characteristics.

2. According to the artificial intelligence-based belt conveyor protection early warning method of claim 1, it is characterized in that: The method for constructing the belt conveyor sensor network to collect belt conveyor operation data and status data comprises: A visual sensing module, a laser scanning module, a thermal imaging module, a sound module and an environmental perception module are set up to form a belt conveyor sensor network; The visual sensing module is composed of high-definition cameras at various positions of the belt conveyor, which is used to obtain image information of the belt conveyor in operation; the laser scanning module is composed of laser scanning sensors, which is used to obtain the three-dimensional point cloud data, contour information and laser beam reflection time of the belt conveyor; the thermal imaging module is composed of infrared thermal imaging sensors, which is used to detect the infrared radiation of the belt conveyor and convert it into temperature data of each component of the belt conveyor; the sound module is composed of sound sensors, which is used to obtain the sound wave signal generated during the operation of the belt conveyor; the sound wave signal includes working sound and noise; The environmental perception module consists of a laser dust sensor, a temperature and humidity sensor, a micro-pressure difference sensor, and a vibration sensor, and is used to obtain the environmental status data of the belt conveyor; The operation data of the belt conveyor is composed of the data collected by the visual sensor module, laser scanning module, thermal imaging module and sound module. The machine status data is obtained by directly extracting the belt conveyor status parameters and operation records, and the status data of the belt conveyor is composed of the machine status data and the environmental status data collected by the environmental perception module.

3. According to the artificial intelligence-based belt conveyor protection early warning method of claim 1, it is characterized in that: The method for processing the operating data to obtain the main modal features and the auxiliary modal features includes: An image processing model is used to process the image information of the belt conveyor operation to obtain the main modal features; the image processing model includes a correction module, a belt appearance module, a belt deviation module, a color detection module, a product module, a safety module and an integration module; The correction module performs denoising, enhancement, pixel correction and color correction operations on the image information of the belt conveyor to obtain a corrected image; the belt appearance module uses an edge detection algorithm to process the corrected image to obtain belt wear information, and uses threshold segmentation and morphological operations to process the corrected image to obtain belt crack information; the belt deviation module uses the RANSAC algorithm to fit the standard edge of the belt, and calculates the standard edge of the belt and the actual edge to obtain the belt deviation data; the color detection module performs color space conversion on the corrected image, and determines the color abnormality information of the equipment or product according to the color threshold; the product module calculates the product stacking situation based on the image depth information, uses the edge detection algorithm and the gray level co-occurrence matrix to extract the product shape features and texture features, and matches the product database according to the product shape features, texture features and color features to obtain product information; the safety module uses a deep learning human body detection model to detect unsafe factors of personnel, and uses a target detection model to detect unsafe factors of equipment and foreign objects around the belt conveyor; the integration module is connected with the correction module, belt appearance module, belt deviation module, color detection module, product module, and safety module, and integrates the information processed by the above modules to output the main modal features of the belt conveyor; The collected data of the laser scanning module, the thermal imaging module and the sound module are processed to obtain auxiliary modal features. The specific method is as follows: the distance data is determined according to the time of laser beam reflection, the belt offset auxiliary data is determined by the distance data, the contour information and the three-dimensional point cloud data are integrated to obtain the current belt conveyor model, and the current belt conveyor model is compared with the standard belt model to obtain the belt wear and deformation auxiliary data; the temperature distribution diagram is drawn according to the temperature data of each component of the belt conveyor, and the thermal change trend is calculated by a statistical method; the vibration signal of the vibration sensor is used to correct the sound wave signal, and the corrected sound wave signal is processed by Fourier transform to identify the working sound and noise, and feature extraction is performed to obtain the operating sound wave feature and the noise feature, and the comprehensive similarity of the combined vector of the operating sound wave feature and the noise feature and the historical sound wave vector of the sound wave database is calculated to obtain the belt conveyor part feature; the sound wave database includes historical sound wave vectors and corresponding part features.

4. According to the artificial intelligence-based belt conveyor protection early warning method of claim 1, it is characterized in that: The method for obtaining the first operating characteristic comprises: Calculate the comprehensive similarity between the auxiliary modal feature vectors, match the auxiliary modal features with the auxiliary modal features according to the highest comprehensive similarity, fuse the two auxiliary modal features according to the matching result to obtain the auxiliary modal correction feature, and combine the auxiliary modal correction feature and the unmatched auxiliary modal feature to form the first auxiliary modal feature; Calculate the comprehensive similarity between the first auxiliary modal feature and the main modal feature, determine the second auxiliary modal feature by the highest comprehensive similarity, calculate the correlation between the main modal feature and the unmatched first auxiliary modal feature, take the unmatched first auxiliary modal feature greater than the correlation threshold as the third auxiliary modal feature, compare the second auxiliary modal feature with the main modal feature to obtain the degree of difference, cross-check the third auxiliary modal feature with the main modal feature to obtain the first main modal feature, perform feature fusion based on the first main modal feature, the second auxiliary modal feature and the degree of difference to obtain the second main modal feature, and form the second main modal feature, the unmatched first auxiliary modal feature and the unmatched main modal feature into a first operating feature; the second auxiliary modal feature and the main modal feature are the same index collected and calculated by different sensing modules; the third auxiliary modal feature is a different index that affects the main modal feature collected and calculated by different sensing modules; The mean, standard deviation and skewness corresponding to the temperature information, acoustic wave characteristics and belt offset in the first operating characteristic are statistically collected, and the temperature stability, acoustic wave stability and offset stability are calculated based on the statistical results. The comprehensive stability is obtained by weighted fusion of various stabilities.

5. According to the artificial intelligence-based belt conveyor protection early warning method of claim 1, it is characterized in that: The method of inputting the state data into a state function to obtain a state factor comprises: The state factors include environmental state factors and machine state factors; Input the environmental state data into the environmental state function to obtain the environmental state factor, the expression is: in is the environmental state factor, is the dust accumulation weight, is the temperature and humidity weight, To measure the impact weight, is the dust concentration, is the current air velocity, is the standard air flow rate, is the current wind pressure, is the standard wind pressure, is the current ambient temperature, is the standard ambient temperature, is the standard deviation of ambient temperature, Current ambient humidity, is the standard ambient humidity, is the standard deviation of ambient humidity, is the indoor light intensity, is the indoor magnetic field strength, is the mean value of mechanical vibration intensity, is the standard mechanical vibration intensity; Input the machine state data into the machine state function to obtain the machine state factor, which is expressed as: in is the machine state factor, For power supply influence weight, is the work efficiency weight, Maintain weights for operations, is the average current of the device, is the maximum rated current of the device, is the current standard deviation, is the maximum rated voltage of the device, is the mean value of the equipment voltage, is the voltage standard deviation, For current work efficiency, For maximum work efficiency, is the current material flow rate, is the base material flow rate, is the current continuous working time of the device, Rated continuous working time of the equipment, is the strip replacement time, Specify the usage time for the strip, is the number of emergency stop operations of the equipment, Manual frequency tuning times for the device. For the strip lubrication interval, Specify lubrication times for the strips.

6. According to the artificial intelligence-based belt conveyor protection early warning method of claim 1, it is characterized in that: The method for obtaining the hidden danger constant comprises: The belt damage hidden danger characteristics and belt tension form a belt quality feature set; the accident quality hidden danger characteristics and machine status factors form an accident quality feature set; the potential hidden danger characteristics, environmental status factors and comprehensive stability form a comprehensive quality feature set; Constructing a belt conveyor depth warning model, the belt conveyor depth warning model includes an input layer, a base model layer and an output layer; the base model layer includes a linear regression base model, a BP neural network base model and a multimodal decision tree base model; The linear regression base model performs performance linear regression on the belt damage hidden danger characteristics and belt tension to predict the first hidden danger constant, uses polynomial feature expansion to process crack information, uses Huber loss function for robustness optimization, uses Adam optimizer to adjust the learning rate, uses belt quality feature set to train the model and verify the model performance; The BP neural network based model performs nonlinear regression on accident quality hidden danger characteristics and machine status factors to predict the second hidden danger constant. The LSTM+Attention mechanism is used to capture the time dependence of accident quality hidden danger characteristics. The cross entropy loss function is used to measure the difference between the probability distribution of the model output and the true distribution. Dynamic regularization is used to automatically adjust the Dropout rate according to the training error. The accident quality feature set is used to train the model and verify the model performance. The multimodal decision tree-based model adopts a multi-head cross-attention mechanism to fuse potential hidden danger features to obtain potential fusion hidden danger features, adopts a decision tree to capture the complex patterns and interactions of potential fusion hidden danger features, environmental state factors and comprehensive stability, performs nonlinear regression to predict the third hidden danger constant, adopts dynamic pruning to adjust the tree depth, adopts weighted classification loss and regression loss to measure model performance, adopts RMSprop optimizer to adjust the learning rate, adopts a comprehensive quality feature set to train the model and verify the model performance; the splitting criterion of the decision tree is specifically as follows: in For the splitting criterion, is the Gini coefficient, is the feature weight, is the interaction weight, is the state weight, is the feature importance, , is the interactive feature among the potential fusion risk features, is the machine state factor; The output layer is connected to the three base models, receiving the prediction results of the base models and outputting the hidden danger constants.

7. According to the artificial intelligence-based belt conveyor protection early warning method of claim 1, it is characterized in that: The method for performing a deep warning according to the hidden danger constant and determining a second warning strategy comprises: The second warning strategy includes performing a second-level warning, performing a third-level warning, and outputting operation safety; the second-level warning and the third-level warning indicate that the first operation characteristics of the same type cumulatively interact with each other and there is a potential risk; When the hidden danger constant is greater than the corresponding hidden danger constant threshold, a second-level warning is issued for the same category of operation; When there is only one hidden danger constant less than or equal to the corresponding hidden danger constant threshold, the ratio of the same type of hidden danger constant to the corresponding hidden danger constant threshold is calculated to obtain the hidden danger constant ratio. If the hidden danger constant ratio is greater than the first ratio threshold, a third-level warning is issued, otherwise the corresponding type of operation is output as safe. When the remaining two hidden danger constants are less than or equal to the corresponding hidden danger constant threshold, the other hidden danger constant is input into the weight function to obtain the hidden danger constant weight, and the hidden danger constant weight and the product of the two hidden danger constants are calculated to obtain the weighted hidden danger constant. If the weighted hidden danger constant is greater than the corresponding hidden danger constant threshold, a third-level warning is issued, otherwise the corresponding class operation safety is output; When the three hidden danger constants are all less than or equal to the corresponding hidden danger constant thresholds, the ratio of each type of hidden danger constant is calculated. If the mean of the hidden danger constant ratio is greater than the second ratio threshold, a third-level warning is issued. Otherwise, the overall operation of the output belt conveyor is safe.

8. A belt conveyor protection warning system based on artificial intelligence, used to execute the method described in any one of claims 1 to 7, characterized in that: include: Data acquisition module: used to build a belt conveyor sensor network to collect belt conveyor operation data and status data; Data processing module: used for processing the operation data to obtain the main modal features and the auxiliary modal features, matching the common features, performing feature correction based on the matching results to obtain the first operation features, and inputting the state data into the state function to obtain the state factor; A primary warning module: used for preliminarily screening the first operation characteristics and determining a first warning strategy according to the preliminarily screening results; Deep warning module: used to construct a belt conveyor deep warning model according to the initial screening results and the state factors, input the operation data and state data of the belt conveyor to be warned into the belt conveyor deep warning model to obtain the hidden danger constant, and perform deep warning according to the hidden danger constant to determine the second warning strategy; Management module: used to manage the operation data, the status data, the preliminary screening results and the hidden danger constants, and to make a first-level warning according to the preliminary screening results and a second-level warning according to the hidden danger constants.

Images