Steel wire rope core conveying belt vulcanization intelligent diagnosis method based on deep learning

Abstract

The invention provides a steel wire rope core conveying belt vulcanization intelligent diagnosis method based on deep learning, and relates to the technical field of conveying belt vulcanization diagnosis. Firstly, temperature sensors and pressure sensors are arranged at multiple points, temperature, pressure and original data in the vulcanization process are collected in real time, and the original data are preprocessed; secondly, reconstructing a temperature field in real time: reconstructing two-dimensional or three-dimensional temperature field distribution of a vulcanization area in real time by utilizing a spatial interpolation algorithm, and carrying out vulcanization effect quantitative calculation: for each sampling point in the temperature field, carrying out integration on the temperature history of the sampling point based on an Arrhenius equation, and calculating an accumulated vulcanization effect value representing the vulcanization degree; an intelligent diagnosis module is started, the calculated temperature field and the accumulated vulcanization effect value are analyzed in real time, and a diagnosis result and early warning information are output; the diagnosis result is displayed in the form of a visual chart and a report in an application layer, and all process data and diagnosis conclusions are archived for quality tracing.

Description

Technical Field

[0001] This invention relates to the field of conveyor belt vulcanization diagnosis technology, and in particular to a deep learning-based intelligent diagnosis method for steel wire rope core conveyor belt vulcanization. Background Technology

[0002] Steel cord conveyor belts are crucial material handling equipment in modern industry, and vulcanization is one of their core manufacturing processes. Vulcanization, performed under high temperature and pressure, causes the rubber molecular chains to cross-link, thus endowing the conveyor belt with excellent physical and mechanical properties. However, traditional vulcanization processes have several drawbacks:

[0003] (1) The process is “black box”: the vulcanization reaction occurs inside the closed vulcanizing machine. The internal temperature field distribution and the degree of vulcanization reaction in each area are difficult to directly observe and quantify. Quality control relies heavily on the experience of the operators.

[0004] (2) Poor quality consistency: Due to uneven temperature distribution of hot plates, pressure fluctuations and other factors, the degree of vulcanization at different locations of the conveyor belt (such as the center and edge, surface and core) often varies, resulting in unstable key properties such as peel strength and adhesive strength of the product, and making it difficult to guarantee the qualified rate of finished products.

[0005] (3) Lack of effective traceability: When quality problems occur, traditional production records only contain simple temperature and pressure setpoints, which cannot provide detailed process data to analyze the root cause, making it difficult to trace the problem and optimize the process;

[0006] (4) Difficulty in optimizing production efficiency and energy consumption: The heat preservation and pressure holding time is usually set based on experience and tends to be conservative, which not only prolongs the production cycle but also causes unnecessary energy waste.

[0007] Therefore, there is an urgent need for a technical solution that can penetrate the "black box" of the vulcanization process to achieve precise quantification of the vulcanization state, intelligent diagnosis of process anomalies, and prediction and control of product quality. Summary of the Invention

[0008] To address the shortcomings of existing technologies, this invention provides a deep learning-based intelligent diagnostic method for vulcanization of steel wire rope conveyor belts. The method aims to solve the following problems: how to accurately and in real-time acquire temperature field distribution data inside the vulcanizing machine (especially in the area covering the conveyor belt); how to establish an accurate quantitative model of the degree of vulcanization (vulcanization effect) based on real-time temperature data; how to intelligently diagnose and provide early warnings for key parameter anomalies, vulcanization uniformity, and vulcanization endpoints during the vulcanization process; and how to achieve traceability and analyzability of the vulcanization quality of each batch of products, providing data-driven decision support for process optimization.

[0009] The specific technical solution of the present invention is as follows:

[0010] On one hand, this invention provides a deep learning-based intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts, comprising the following steps:

[0011] Step 1: Arrange temperature and pressure sensors at multiple points to collect temperature, pressure, and raw data in real time during the vulcanization process, and preprocess the raw data.

[0012] The temperature sensor and pressure sensor are installed in the sensor slots; the sensor slots are symmetrically arranged on the upper and lower surfaces of each frame in the vulcanizing hot plate; two sensor slots on the same frame are respectively arranged with one temperature sensor and one pressure sensor; the sensor arrangement of adjacent frames is opposite; the sensors of the temperature sensor group and the pressure sensor group are arranged in an alternating manner on both sides of the upper and lower vulcanizing hot plates.

[0013] The pressure sensor is also installed on the main hydraulic cylinder and air bladder pipeline of the vulcanizing machine;

[0014] The preprocessing specifically includes filtering and outlier removal;

[0015] Step 2: Real-time temperature field reconstruction: The two-dimensional or three-dimensional temperature field distribution of the vulcanization region is reconstructed in real time using a spatial interpolation algorithm;

[0016] Specifically, based on the real-time temperature values ​​from the temperature sensors, a temperature field distribution map covering the entire effective area grid of the conveyor belt is calculated and updated at set intervals.

[0017] Step 3: Quantitative calculation of sulfidation effect: For each sampling point in the temperature field, based on the Arrhenius equation, the temperature history of the sampling point is integrated to calculate the cumulative sulfidation effect value characterizing the degree of sulfidation.

[0018] Specifically, based on the temperature history T(t) of the sampling point since the start of sulfidation, the cumulative sulfidation effect CET is calculated using the integral form of the following Arrhenius equation:

[0019] ;

[0020] In the formula, E a T is the activation energy for the sulfidation reaction. ref Let t be the reference temperature, R be the ideal gas constant, and t be the reference temperature. 始 The initial state time of the vulcanization reaction, i.e., the start time of vulcanization, is t. 终 The time point at which the vulcanization reaction reaches the specified process is the vulcanization end time.

[0021] Step 4: Multi-dimensional intelligent diagnosis: Activate the intelligent diagnosis module to perform real-time analysis on the calculated temperature field and cumulative sulfurization effect value, and output diagnosis results and early warning information;

[0022] Specifically, an intelligent diagnostic module is established by combining isolated forest with threshold rules. Abnormal samples are isolated by randomly segmenting the feature space, and abnormal working conditions are identified in real time during the sulfurization process.

[0023] Step 5: Results Presentation and Traceability: Display diagnostic results in the form of visual charts and reports at the application layer, and archive all process data and diagnostic conclusions for quality traceability.

[0024] On the other hand, the intelligent diagnostic method for vulcanization of steel wire rope core conveyor belt based on deep learning is implemented through the following system: including sensors, PLC, and industrial control computer;

[0025] The sensors include temperature sensors and pressure sensors, which are installed on the hot plate and key areas of the vulcanizing machine.

[0026] The PLC is equipped with an edge computing node to preprocess sensor data and execute a local real-time alarm on the PLC terminal when the data is abnormal, and upload the alarm data to the industrial control computer server in ModBus-TCP structure.

[0027] The industrial control computer includes a data storage module, a core algorithm engine module, and an intelligent diagnostic module. The data storage module stores historical process data, including vulcanization time-vulcanization temperature and vulcanization time-vulcanization pressure data. The core algorithm engine module reconstructs the temperature field and quantifies the vulcanization effect. The intelligent diagnostic module executes diagnostic tasks based on the output of the core algorithm engine module. The industrial control computer platform provides a human-machine interface for real-time monitoring, historical data tracing, diagnostic report generation, and process parameter management. The diagnostic report includes vulcanization process curves and vulcanization result predictions.

[0028] The diagnostic tasks include process anomaly diagnosis, vulcanization uniformity diagnosis, vulcanization endpoint prediction diagnosis, and comprehensive batch quality assessment.

[0029] The process anomaly diagnosis is performed by setting process procedure rules and using existing LSTM models in combination with existing Mamba models to learn the vulcanization time-vulcanization temperature and vulcanization time-vulcanization pressure curve patterns under normal operating conditions in order to identify minor anomalies.

[0030] The vulcanization uniformity diagnosis is made by calculating the difference index of the cumulative vulcanization effect value (CET) between the center point of the temperature field and the four edge points, and then judging based on a set threshold.

[0031] The sulfurization endpoint prediction and diagnosis involves diagnosing process anomalies after the sulfurization enters the sulfurization heat preservation stage. It establishes a complementary relationship between "physical field → quantitative index" by combining the CET values ​​of all current points and the Fourier thermal conductivity model to predict the time to reach the expected degree of sulfurization, and at the same time diagnoses the bottleneck points that determine the completion of sulfurization.

[0032] The comprehensive quality assessment of the batch is conducted after the vulcanization process is completed. Based on the complete data of the entire vulcanization process, a comprehensive quality score is obtained by multiplying the weights and data results using a weighted scoring method. The batch is then rated based on the comprehensive quality score.

[0033] The beneficial effects of adopting the above technical solution are as follows:

[0034] This invention provides a deep learning-based intelligent diagnostic method for vulcanization of steel wire rope conveyor belts, which specifically includes the following beneficial effects:

[0035] (1) Process transparency and quantification: For the first time, the invisible degree of sulfidation is accurately quantified and visualized through the sulfidation effect (CET) value, turning the "black box" operation into a transparent and controllable digital process.

[0036] (2) Significantly improve quality consistency: Through real-time diagnosis of vulcanization uniformity, the vulcanization difference between the center and the edge can be detected and warned in time, guiding process adjustment, thereby greatly improving the stability and consistency of product quality.

[0037] (3) Realize proactive intelligent diagnosis: Change passive quality inspection to proactive process diagnosis, which can detect problems such as abnormal heating rate, pressure fluctuation and uneven uniformity in real time during vulcanization, prevent problems before they occur and reduce scrap rate.

[0038] (4) Optimize production efficiency and energy consumption: Based on the accurate prediction of the vulcanization endpoint, unnecessary over-vulcanization can be avoided, the vulcanization time can be shortened safely, thereby improving equipment turnover and reducing unit energy consumption.

[0039] (5) Provide strong data support: Complete digital records provide a solid foundation for quality traceability, correlation analysis between process parameters and product performance, and process optimization based on big data, thus promoting the digital transformation of enterprises. Attached Figure Description

[0040] Figure 1 Schematic diagram of the intelligent diagnostic system architecture according to an embodiment of the present invention;

[0041] In the diagram, 1-steel wire core conveyor belt, 2-lower frame, 3-lower vulcanizing hot plate, 4-upper vulcanizing hot plate, 5-upper frame, 6-pressure bolt, 7-pressure nut;

[0042] Figure 2Schematic diagram of sensor installation according to an embodiment of the present invention;

[0043] In the diagram, 8-pressure sensor, 9-temperature sensor, 10-temperature sensor mount;

[0044] Figure 3 Enlarged view of sensor installation according to an embodiment of the present invention;

[0045] Figure 4 Overall flowchart of the intelligent diagnostic method according to an embodiment of the present invention;

[0046] Figure 5 Flowchart of the diagnostic prediction architecture of the intelligent diagnostic method according to an embodiment of the present invention;

[0047] Figure 6 A schematic diagram of the visual diagnostic interface in an embodiment of the present invention. Detailed Implementation

[0048] The specific implementation methods of this application will be further described in detail below with reference to the accompanying drawings and embodiments.

[0049] Example 1:

[0050] On one hand, this invention provides a deep learning-based intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts, comprising the following steps:

[0051] Step 1: Deploy high-precision temperature and pressure sensors at multiple points to collect temperature, pressure, and raw data during the vulcanization process in real time, and preprocess the raw data.

[0052] The temperature sensor and pressure sensor are installed in the sensor slots; the sensor slots are symmetrically arranged on the upper and lower surfaces of each frame in the vulcanizing hot plate; two sensor slots on the same frame are respectively arranged with one temperature sensor and one pressure sensor; the sensor arrangement of adjacent frames is opposite; the sensors of the temperature sensor group and the pressure sensor group are arranged in an alternating manner on both sides of the upper and lower vulcanizing hot plates.

[0053] The pressure sensor is also installed on the main hydraulic cylinder and air bladder pipeline of the vulcanizing machine;

[0054] The preprocessing specifically includes filtering and outlier removal;

[0055] Step 2: Real-time temperature field reconstruction: The two-dimensional or three-dimensional temperature field distribution of the vulcanization region is reconstructed in real time using a spatial interpolation algorithm;

[0056] Specifically, based on the real-time temperature values ​​from the temperature sensors, a temperature field distribution map covering the entire effective area grid of the conveyor belt is calculated and updated at set intervals.

[0057] Step 3: Quantitative calculation of sulfidation effect: For each sampling point in the temperature field, based on the Arrhenius equation, the temperature history of the sampling point is integrated to calculate the cumulative sulfidation effect value (Cure Equivalent Time, CET) that characterizes the degree of sulfidation.

[0058] Specifically, based on the temperature history T(t) of the sampling point since the start of sulfidation, the cumulative sulfidation effect CET is calculated using the integral form of the following Arrhenius equation:

[0059] ;

[0060] In the formula, E a T is the activation energy for the sulfidation reaction. ref Let t be the reference temperature, R be the ideal gas constant, and t be the reference temperature. 始 The initial state time of the vulcanization reaction, i.e., the start time of vulcanization, is t. 终 The time point at which the vulcanization reaction reaches the specified process is the vulcanization end time.

[0061] Step 4: Multi-dimensional intelligent diagnosis: Activate the intelligent diagnosis module to perform real-time analysis on the calculated temperature field and cumulative sulfurization effect value, and output diagnosis results and early warning information;

[0062] Specifically, an intelligent diagnostic module is established by combining isolated forest with threshold rules. Abnormal samples are isolated by randomly segmenting the feature space, and abnormal working conditions are identified in real time during the sulfurization process.

[0063] Step 5: Results Presentation and Traceability: Display diagnostic results in the form of visual charts and reports at the application layer, and archive all process data and diagnostic conclusions for quality traceability.

[0064] On the other hand, the intelligent diagnostic method for vulcanization of steel wire rope core conveyor belt based on deep learning is implemented through the following system: including sensors, PLC, and industrial control computer;

[0065] The sensors include temperature sensors and pressure sensors, which are installed on the hot plate and key areas of the vulcanizing machine.

[0066] The PLC is equipped with an edge computing node to preprocess sensor data and execute a local real-time alarm on the PLC terminal when the data is abnormal, and upload the alarm data to the industrial control computer server in ModBus-TCP structure.

[0067] The industrial control computer includes a data storage module, a core algorithm engine module, and an intelligent diagnostic module. The data storage module persistently stores historical process data, including vulcanization time-vulcanization temperature and vulcanization time-vulcanization pressure data. The core algorithm engine module reconstructs the temperature field and quantifies the vulcanization effect. The intelligent diagnostic module executes diagnostic tasks based on the output of the core algorithm engine module. The industrial control computer platform provides a human-machine interface for real-time monitoring, historical data tracing, diagnostic report generation, and process parameter management. The diagnostic report includes vulcanization process curves and vulcanization result predictions.

[0068] The diagnostic tasks include process anomaly diagnosis, vulcanization uniformity diagnosis, vulcanization endpoint prediction diagnosis, and comprehensive batch quality assessment.

[0069] The process anomaly diagnosis is performed by setting process procedure rules and using existing LSTM models in combination with existing Mamba models to learn the vulcanization time-vulcanization temperature and vulcanization time-vulcanization pressure curve patterns under normal operating conditions, so as to identify minor anomalies and achieve early warning.

[0070] The vulcanization uniformity diagnosis is made by calculating the difference index of the cumulative vulcanization effect value (CET) between the center point of the temperature field and the four edge points, and then judging based on a set threshold.

[0071] The sulfurization endpoint prediction and diagnosis involves diagnosing process anomalies after the sulfurization enters the sulfurization heat preservation stage. It establishes a complementary relationship between "physical field → quantitative index" by combining the CET values ​​of all current points and the Fourier thermal conductivity model to predict the time to reach the expected degree of sulfurization, and at the same time diagnoses the bottleneck points that determine the completion of sulfurization.

[0072] The comprehensive quality assessment of the batch is conducted after the vulcanization process is completed. Based on the complete data of the entire vulcanization process, a weighted scoring method (the weights can be set) is used to output the product of the weights and the data results to obtain a comprehensive quality score. The comprehensive quality score is then used to assign a rating, providing a basis for quality decision-making.

[0073] Example 2:

[0074] like Figure 1 As shown, the steel wire core conveyor belt 1 is sandwiched between the lower vulcanizing hot plate 3 and the upper vulcanizing hot plate 4. The upper and lower vulcanizing hot plates are pressurized by the pressure bolts 6 and pressure nuts 7 on the upper frame 5 and the lower frame 2. In this embodiment, universal sensor slots for mounting pressure sensors 8 and temperature sensors 9 are provided on the surfaces of the upper and lower frames that contact the vulcanizing plates (for ease of installation, universal sensor slots are machined on both the upper and lower surfaces of each frame), which provides good interchangeability between the upper and lower frames. The reference numeral 10 is the temperature sensor seat 10. The universal sensor slots are symmetrically arranged on both sides of the upper and lower frames, and their actual positions are determined by the size of the conveyor belt being vulcanized.

[0075] Two sensor slots on the same frame are respectively equipped with a temperature sensor and a pressure sensor; for example Figure 2 , Figure 3 As shown, the sensors on two adjacent frames are arranged in opposite ways; the sensors on two frames clamped by pressure bolts and nuts are also arranged in opposite ways. That is, the sensor units of the temperature sensor group and the pressure sensor group are arranged in an alternating manner on both sides of the upper and lower vulcanizing hot plates.

[0076] In this embodiment, a K-type armored explosion-proof temperature sensor and a PVDF pressure sensor are embedded in the sensor slot. High-temperature resistant pressure sensors are installed on the main hydraulic cylinder and air bladder pipeline of the vulcanizing machine. All sensor signals use standard 4-20mA or RS485 signals.

[0077] Siemens S7-1200 series PLCs are used as edge computing nodes. The PLCs are responsible for high-speed acquisition of data from various sensors, executing filtering algorithms, and triggering real-time local alarms when temperature or pressure exceeds safe limits. Processed data is uploaded via the Profinet industrial Ethernet protocol and deployed on a local server. The sensor time-series data is stored using the Influx DB time-series database, while relational data such as process recipes and diagnostic reports are stored using MySQL. The core component is the intelligent diagnostic module, developed in Python and integrated with deep learning frameworks such as TensorFlow.

[0078] The human-computer interaction interface in this embodiment is developed based on Web technology and can be accessed by users through a browser. The front end uses the Vue.js framework, and the back end interacts with the server via API.

[0079] Example 3:

[0080] like Figure 4 As shown, the specific steps of the intelligent diagnostic method of the present invention are as follows:

[0081] Step S1: Data Acquisition and Preprocessing. The PLC acquires data from all sensors at a frequency of 100Hz. An adaptive wavelet threshold denoising algorithm is used to eliminate noise in the temperature data; the temperature time series data is then constructed into a time series dataset.

[0082] (1)

[0083] In formula (1) x i This represents the temperature value at the i / 100th second, where i is the sample value.

[0084] The time series data x is decomposed into low-frequency approximation coefficients A corresponding to the temperature change trend using wavelet decomposition. nAnd the high-frequency detail coefficients D1-D5 corresponding to the noise; calculate the standard deviation σ of D1-D5. j j=[1,2,3,4,5];

[0085] (2)

[0086] In formula (2), d i Elements in D1-D5;

[0087] Calculate σ j Dynamic threshold λ j :

[0088] (3)

[0089] For the high-frequency coefficients of each layer from D1 to D5, a threshold function T is defined. λ (d i );

[0090] (4)

[0091] Define Stein's unbiased risk estimation as minimizing the denoised error R(λ) in real time:

[0092] (5)

[0093] In formula (5), I() is an indicator function. It takes the value 1 when the condition is true and 0 otherwise. The optimal threshold is the λ value that minimizes R(λ).

[0094] Using "missing data point t" k Centered on i, establish 5 sampling points t1, ...,t2, ...,t3, with i = [1, 2, ..., 5]. i A sliding window of t5 is used, employing Gaussian kernel function for locally weighted linear regression, while ensuring compliance with normalization conditions.

[0095] (6)

[0096] w in formula (6) i The weights are locally linearly weighted, and σ is the standard deviation of t1 to t5;

[0097] Then, a linear model is fitted using the weighted least squares method to construct T. k =at i +b completes data loss caused by sensor disconnection;

[0098] (7)

[0099] In formula (7), a is the slope of the linear fit, b is the intercept of the linear fit, and S is the sum of squares of the deviations of the fitted function;

[0100] Simultaneously, outliers caused by transient disturbances are detected and removed using the Grubbs criterion to obtain the completed sequence T. completed :

[0101] (8)

[0102] Calculate T completed Given the mean μ and the unbiased standard deviation λ, calculate the Grubbs statistic G;

[0103] (9)

[0104] Based on the sample size n (n=5) and significance level α s (Typically, 0.05 is used, i.e., 95% confidence level), cyclic decision T completed If each element in the middle, , then t i Outliers are removed; otherwise, the data is considered normal.

[0105] (10)

[0106] In formula (10) This is the critical value of the t-distribution;

[0107] Step S2: Dynamic Temperature Field Reconstruction: Invoking the interpolation algorithm based on Radial Basis Function (RBF):

[0108] (11)

[0109] In formula (11) It is a univariate function, t=(t1,t2,…,t…). 14 (c1, c2, ..., c) is any point in d-dimensional space (i.e., the interpolation variable); 14 The center point (coinciding with known data points) is based on the radial basis function (RBF); the distance metric in space uses Euclidean distance.

[0110] (12)

[0111] (13)

[0112] In formula (13), j represents the j-th sensor, and k represents the dimension of the space. These are the weighting coefficients based on radial basis functions. p represents the weighting coefficients of the polynomial terms. k(x) is a polynomial basis in k-dimensional space, defined as d=2, p1=1, p2=x, p3=y, corresponding to constant terms and linear terms.

[0113] Dynamic temperature field reconstruction is based on the real-time temperature values ​​from temperature sensors (data from 14 temperature sensors) to calculate a 100×50 grid temperature field distribution map covering the entire effective area of ​​the conveyor belt, and it is updated every 5 seconds.

[0114] Step S3: Quantitative calculation of the sulfidation effect; key points in the temperature field are selected (such as the center point and the four corner points). For each key point, the cumulative sulfidation effect CET is calculated using the integral form of the following Arrhenius equation, based on its temperature history T(t) since the start of sulfidation:

[0115] (14)

[0116] In formula (14), the activation energy E of the sulfidation reaction is... a and reference temperature T ref This is a constant obtained by calibration on the vulcanizer based on the specific rubber compound formulation. R is the ideal gas constant. This integral calculation is performed simultaneously with the dynamic temperature field reconstruction process in step S2.

[0117] Step S4: Multi-dimensional intelligent diagnosis, such as Figure 5 As shown, an intelligent diagnostic module is established using a fusion scheme of "isolated forest" and "threshold rule". By randomly segmenting the feature space, "abnormal samples" are quickly isolated, and anomalies such as "uneven temperature, pressure leakage, and steel wire core displacement" are identified in real time during the vulcanization process.

[0118] Specifically, an isolated forest containing 100 binary iTrees was constructed, and the average path length of temperature t and pressure data p in the iTree forest was calculated. :

[0119] (15)

[0120] In formula (15) h i (t) represents the number of edges that temperature t passes through from the leaf node to the root node on the i-th i-Tree;

[0121] To eliminate the influence of sample size on path length, a normalized score is introduced for each iTree with a sample size of n, as shown in the following formula:

[0122] (16)

[0123] In formula (16), c(n) is the expected average path length of normal samples when the number of samples is n, used for standardization:

[0124] (17)

[0125] In formula (17), H(n-1) is the (n-1)th harmonic number, approximately ln(k) + 0.5772, which is Euler's constant.

[0126] If the normalized score s(t,n)≈1: the average path length of sample t is much less than the expected value c(n) of normal samples, and it is judged as an abnormal sample; if s(t,n)≈0.5: the average path length of sample t is close to c(n), and it is judged as a normal sample; if s(t,n)≈0: the average path length of sample t is much greater than c(n), and it is judged as a normal sample.

[0127] The analytic hierarchy process (AHP) is used to determine the weights based on the importance of features such as vulcanization temperature, pressure, and time. A Bayesian network structure is constructed, and fault cause nodes and abnormal feature nodes are designed. A conditional probability table for "parent node → child node" is determined based on historical fault data and maximum likelihood estimation.

[0128] The posterior probability of each fault cause node is calculated using Bayes' theorem, and the one with the highest posterior probability is the most likely root cause.

[0129] The LSTM+Mamba network is trained using a parallel feature fusion method. The input sequence is fed into the LSTM network and the Mamba network respectively. The extracted features are fused by splicing and intention mechanism, and then the prediction result of the disulfide process is output through a fully connected layer.

[0130] Using the Adam optimizer, the loss function is specified as mean squared error (MSE):

[0131] (18)

[0132] In formula (18), For predicted values, The true value is n, and the sample size is n.

[0133] Specify learning rate First-order moment exponential decay rate The exponential decay rate of the second moment Numerical stability term Update the parameters using the following steps:

[0134] Update first-order moment estimate :

[0135] (19)

[0136] in For gradient;

[0137] Update the second moment estimate :

[0138] (20)

[0139] Deviation corrections are applied to the first and second moments:

[0140] (twenty one)

[0141] (twenty two)

[0142] Update parameters :

[0143] (twenty three)

[0144] In formula (23) The core of the formula is: subtract the "adjusted step size" from the old parameters to obtain the new parameters.

[0145] By using the "early stopping method" to avoid overfitting, the final model's overall error is less than 3%, meeting the accuracy requirements for diagnosis and prediction. The following table shows the Python configuration to achieve early stopping.

[0146] Table 1. Parameter settings for the early cessation method:

[0147] Step S5: Result Presentation and Traceability: Diagnostic and predictive results (such as "low heating rate" and "uniformity deviation warning") are pushed to the application layer interface in real time and alerted to the on-site operator via audible and visual alarms. After vulcanization, the system automatically generates a PDF report containing all process curves, diagnostic events, and the final quality assessment, and stores it in association with the production batch number.

[0148] The core technology of this invention is the intelligent diagnostic module, which integrates multiple algorithm models to conduct a comprehensive health assessment of the vulcanization process. It mainly includes four sub-functions: process anomaly diagnosis, which uses rule-based diagnosis based on set process specifications and employs an LSTM+Mamba model to learn normal operating condition curve patterns to identify minor anomalies and achieve early warning; vulcanization uniformity diagnosis, which calculates the difference in CET values ​​between the temperature field center point and the four edge points, and makes judgments based on set warning and alarm thresholds; vulcanization endpoint prediction diagnosis, which activates a model-based predictive control algorithm during the heat preservation stage, combining the current CET value and the heat conduction model to predict the time to reach the target, and simultaneously diagnoses the bottleneck points that determine the completion of vulcanization; and batch quality comprehensive evaluation, which, based on complete data after vulcanization, outputs a comprehensive quality score and rating through a weighted scoring model, providing a basis for quality decisions.

[0149] The industrial PC's human-machine interface provides a powerful visual diagnostic interface, such as... Figure 6 This interface displays real-time curves and diagnostic status of key point temperatures, curing effect (CET), and the overall temperature. It intuitively presents intelligent diagnostic results, primarily including real-time multi-curve display (showing real-time temperature curves at the center and edge of the hot plate), dynamic cumulative CET curves (reflecting the curing reaction progress during the insulation stage), target and threshold lines (including target CET levels and uniformity difference alarm threshold ranges), and intelligent diagnostic event annotations (automatically adding warning markers such as "uniformity exceeds tolerance" at specific points on the timeline). Through this interface, engineers and operators can intuitively grasp the curing status, identify potential problems, and predict the endpoint, achieving intelligent and visualized production management.

[0150] Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a computer program product.

[0151] The various embodiments in this application are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0152] The scope of protection of this application is not limited to the embodiments described above. Obviously, those skilled in the art can make various modifications and variations to this disclosure without departing from the scope and spirit of this disclosure. If such modifications and variations fall within the scope of this solution and its equivalents, then the intent of this disclosure also includes these modifications and variations.

Claims

1. A deep learning-based intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts, characterized in that, Includes the following steps: Step 1: Arrange temperature and pressure sensors at multiple points to collect temperature, pressure, and raw data during the vulcanization process in real time, and preprocess the raw data; the preprocessing specifically includes filtering and outlier removal; Step 2: Real-time temperature field reconstruction: The two-dimensional or three-dimensional temperature field distribution of the vulcanization region is reconstructed in real time using a spatial interpolation algorithm; Step 3: Quantitative calculation of sulfidation effect: For each sampling point in the temperature field, based on the Arrhenius equation, the temperature history of the sampling point is integrated to calculate the cumulative sulfidation effect value characterizing the degree of sulfidation. Step 4: Multi-dimensional intelligent diagnosis: Activate the intelligent diagnosis module to perform real-time analysis on the calculated temperature field and cumulative sulfurization effect value, and output diagnosis results and early warning information; Step 5: Results Presentation and Traceability: Display diagnostic results in the form of visual charts and reports at the application layer, and archive all process data and diagnostic conclusions for quality traceability.

2. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning according to claim 1, characterized in that, The temperature sensor and pressure sensor mentioned in step 1 are installed in the sensor slots; the sensor slots are symmetrically arranged on the upper and lower surfaces of each frame in the vulcanizing hot plate; two sensor slots on the same frame are respectively arranged with a temperature sensor and a pressure sensor; the sensor arrangement of adjacent frames is opposite; the sensors of the temperature sensor group and the pressure sensor group are arranged alternately on both sides of the upper and lower vulcanizing hot plates.

3. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning according to claim 2, characterized in that, The pressure sensor is also installed on the main hydraulic cylinder and air bladder pipeline of the vulcanizing machine.

4. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning according to claim 1, characterized in that, Step 2 specifically calculates a temperature field distribution map covering the entire effective area grid of the conveyor belt based on the real-time temperature value of the temperature sensor, and updates it once every set time interval.

5. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning according to claim 1, characterized in that, Step 3 specifically involves: calculating the cumulative sulfidation effect CET based on the temperature history T(t) of the sampling point since the start of sulfidation, using the integral form of the following Arrhenius equation: ； In the formula, E a T is the activation energy for the sulfidation reaction. ref Let t be the reference temperature, R be the ideal gas constant, and t be the reference temperature. 始 The initial state time of the vulcanization reaction, i.e., the start time of vulcanization, is t. 终 The time point at which the vulcanization reaction reaches the specified process is the vulcanization end time.

6. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning according to claim 1, characterized in that, In step 4, an intelligent diagnostic module is established by using a fusion scheme of isolated forest and threshold rules. Abnormal samples are isolated by randomly segmenting the feature space, and abnormal working conditions are identified in real time during the sulfidation process.

7. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning as described in claim 1 is implemented through an intelligent diagnostic system for vulcanization of steel wire rope core conveyor belts based on deep learning, characterized in that... Including sensors, PLCs, and industrial computers; The sensors include temperature sensors and pressure sensors, which are installed on the hot plate and key areas of the vulcanizing machine. The PLC is equipped with an edge computing node to preprocess sensor data and execute a local real-time alarm on the PLC terminal when the data is abnormal, and upload the alarm data to the industrial control computer server in ModBus-TCP structure. The industrial control computer includes a data storage module, a core algorithm engine module, and an intelligent diagnostic module. The data storage module stores historical process data, including vulcanization time-vulcanization temperature and vulcanization time-vulcanization pressure data. The core algorithm engine module reconstructs the temperature field and quantifies the vulcanization effect. The intelligent diagnostic module executes diagnostic tasks based on the output of the core algorithm engine module. The industrial control computer platform provides a human-machine interface for real-time monitoring, historical data tracing, diagnostic report generation, and process parameter management. The diagnostic report includes vulcanization process curves and vulcanization result predictions.

8. The intelligent diagnostic method for vulcanization of steel wire rope core conveyor belts based on deep learning according to claim 7, characterized in that, The diagnostic tasks include process anomaly diagnosis, vulcanization uniformity diagnosis, vulcanization endpoint prediction diagnosis, and comprehensive batch quality assessment. The process anomaly diagnosis is performed by setting process procedure rules and using existing LSTM models in combination with existing Mamba models to learn the vulcanization time-vulcanization temperature and vulcanization time-vulcanization pressure curve patterns under normal operating conditions in order to identify minor anomalies. The vulcanization uniformity diagnosis is made by calculating the difference index of the cumulative vulcanization effect value (CET) between the center point of the temperature field and the four edge points, and then judging based on a set threshold. The vulcanization endpoint prediction and diagnosis involves diagnosing process anomalies after vulcanization enters the vulcanization heat preservation stage. It establishes a complementary relationship between "physical field → quantitative index" by combining the CET values ​​of all current points and the Fourier thermal conductivity model to predict the time to reach the expected degree of vulcanization, and at the same time diagnoses the bottleneck points that determine the completion of vulcanization. The comprehensive quality assessment of the batch is conducted after the vulcanization process is completed. Based on the complete data of the entire vulcanization process, a comprehensive quality score is obtained by multiplying the weights and data results using a weighted scoring method. The batch is then rated based on the comprehensive quality score.

Images