Vertical furnace body fault diagnosis system and method

Abstract

The invention provides a vertical furnace body fault diagnosis system and method, and belongs to the technical field of fault diagnosis, and the system comprises a state diagram construction module which is used for carrying out the real-time monitoring of each working part in a furnace body through a plurality of measurement devices, and sequentially constructing the measurement state diagram of each working part at each moment according to the time sequence; the image preprocessing module is used for collecting a working state diagram in the furnace body and carrying out image preprocessing on the collected working state diagram; the comparative analysis module is used for carrying out comparative analysis on the preprocessed image and the first standard image, calculating the difference between the preprocessed image and the first standard image in the aspect of image features, and meanwhile, carrying out abnormal comparison on the second standard image of the working part and the measurement state image; and the fault diagnosis module is used for determining the characteristic quantity of the working part according to the difference and the abnormal comparison result, performing fault diagnosis on the working part according to the mapping relation, locking the final fault type and performing early warning, and realizing accurate and advanced diagnosis of the furnace body fault of the vertical furnace.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of fault diagnosis, in particular to a vertical furnace body fault diagnosis system and method. BACKGROUND

[0002] The vertical furnace body is widely used in integrated circuit wafer production, and the normal operation of its working components such as heating elements and thermocouples is crucial to production efficiency and product quality. However, the existing vertical furnace body fault diagnosis method is simple and rough, and has many shortcomings, such as insufficient comprehensive monitoring of working components, delayed response after fault occurs, and the like. Therefore, the accuracy and timeliness of vertical furnace body fault diagnosis need to be improved.

[0003] Therefore, the present application provides a vertical furnace body fault diagnosis system and method. SUMMARY

[0004] The present application provides a vertical furnace body fault diagnosis system and method to solve the above technical problems.

[0005] The present application provides a vertical furnace body fault diagnosis system, comprising: A state diagram construction module for real-time monitoring of each working component in the furnace body by using a plurality of measurement devices pre-configured and installed on the vertical furnace body, and sequentially constructing a measurement state diagram of each working component at each time according to time sequence; An image preprocessing module for acquiring a working state diagram inside the furnace body based on an image acquisition device arranged inside the vertical furnace body, and performing image preprocessing on the acquired working state diagram, the preprocessing including image filtering, noise reduction, contrast enhancement, and edge detection; A comparative analysis module for comparing and analyzing each preprocessed working state diagram with a first standard diagram of the vertical furnace under the corresponding set operating mode, calculating the difference in image features between the two, and at the same time, comparing and analyzing the working component under the set operating mode according to a second standard diagram and the constructed measurement state diagram; A fault diagnosis module for determining the characteristic quantity of the working component according to the difference between the working state diagram and the first standard diagram and the abnormal comparison result, and locking the final fault type and issuing a warning for the working component according to the pre-established mapping relationship between the characteristic quantity and the fault type.

[0006] Preferably, the state diagram construction module comprises: A mode determination unit for determining the configuration and installation mode of the measurement device according to the type and monitoring requirements of the working component, wherein temperature sensors, current sensors and voltage sensors are uniformly arranged on the surface and around the heating element, and temperature sensors and vibration sensors are arranged at the measurement end and connection end of the thermocouple, respectively; A model establishing unit is configured to establish a sensor data space-time calibration model to eliminate clock bias among different sensors based on cross-correlation analysis. A detection and repair unit is configured to detect and repair outliers in measurement results by using a 3σ criterion combined with a moving average filter. A correlation analysis unit is configured to construct a parameter correlation matrix, analyze nonlinear correlations among parameters through correlation coefficients, eliminate redundant parameters, and retain core characteristic parameters. A visualization unit is configured to perform dimensionality reduction visualization on multi-dimensional parameters by using a T-SNE algorithm, represent third-dimensional parameter values in a two-dimensional coordinate system with a color gradient, and form a three-dimensional measurement state diagram containing real-time parameter values, change trends, and correlation relationships, wherein the measurement state diagram covers various key parameter information of a work component and change trends of each key parameter over time.

[0007] Preferably, the method further comprises: A graph construction module is configured to, when a vertical furnace is in a normal running state in a set running mode and various performance indicators reach optimal values corresponding to the set running mode, collect work state graphs of the furnace body in multiple time periods, and regard the graphs as first standard graphs. A standard determination module is configured to pre-process and extract features of the first standard graphs to obtain image feature vectors in the normal running state, and determine characteristic parameters and standard ranges of the first standard graphs by statistical analysis and machine learning algorithms.

[0008] Preferably, the fault diagnosis module comprises: A comparison unit is configured to compare each key parameter in the measurement state diagram with a standard range in a second standard graph to determine whether the corresponding key parameter is out of a normal range. A coincidence judgment unit is configured to analyze whether correlation relationships among the key parameters conform to standard correlation relationships specified in the second standard graph, and compare image features in the measurement state diagram with image features in the second standard graph to analyze whether the work component has undergone abnormal changes. A feature quantity acquisition unit is configured to combine differences between the work state graph and the first standard graph, fuse processing of each comparison result by using a comprehensive analysis algorithm, and determine feature quantities of abnormal conditions of the work component running state, wherein the feature quantities further include: a duration of abnormality of each key parameter, a frequency of abnormality, and propagation and correlation of abnormality among different parameters and image features.

[0009] Preferably, the fault diagnosis module further comprises: The nonlinear mapping unit is configured to collect historical fault data of the working part of the shaft furnace, wherein the historical fault data comprises characteristic quantity data under different fault types, and a nonlinear mapping model between the characteristic quantity data and the fault types is established by using a machine learning algorithm to train and learn the characteristic quantity data; The fault locking unit is configured to input the determined characteristic quantity into the established mapping model during the fault diagnosis process, output a possible fault type and a probability of the possible fault type, and lock a final fault type according to a preset probability threshold and a fault diagnosis rule. The mode formulating unit is configured to formulate different warning levels and warning modes according to the severity of the fault type and the possible influence.

[0010] Preferably, the warning levels and warning modes comprise: When the fault is slight, a low-intensity sound warning and a system interface prompt warning are adopted; When the fault is serious, a high-intensity sound warning, a flashing light warning, a short message warning and an email warning are sent to relevant personnel, and corresponding emergency handling measures are automatically triggered.

[0011] Preferably, the method further comprises: The self-learning module is configured to input new fault data and diagnosis results into the self-learning module when a new fault type is diagnosed or it is found that the existing mapping relationship has an error, and adjust and optimize the first standard graph, the second standard graph and the mapping relationship by using a machine learning algorithm.

[0012] Preferably, the characteristic quantity acquisition unit comprises: The first extraction sub-block is configured to extract a difference characteristic vector of the working state graph and the first standard graph in a gray space wherein, represents the i-th difference characteristic; n represents the total number of difference characteristics; The second extraction sub-block is configured to extract a difference characteristic vector of the temperature field distribution wherein, represents the temperature deviation of the j-th temperature sampling point; m represents the total number of temperature sampling points; The third extraction sub-block is configured to extract a difference characteristic vector of the texture feature wherein, represents the difference percentage of the u-th texture feature; k represents the number of texture features; The construction sub-block is configured to construct a time sequence weight matrix for the difference characteristic vectors of the continuous N time points. The distribution computing sub-block is configured to construct a dynamic Bayesian fusion model based on a Markov chain Monte Carlo method and in combination with difference feature vectors of N time points and a time sequence weight matrix, and calculate a posterior probability distribution through a Gibbs sampling algorithm; The feature fusion sub-block is configured to construct a fuzzy neural network based on the posterior probability distribution to perform feature fusion, wherein a three-layer fuzzy neural network is designed, the input layer is the difference feature vector, the hidden layer adopts a Gaussian membership function, and the output layer adopts a weighted summation function; The feature quantity determining sub-block is configured to calculate a comprehensive abnormality index, a time evolution trend of the feature quantity, and a spatial distribution gradient of the feature quantity according to the feature fusion result, and determine the feature quantity of the abnormal operation state of the work component;

[0013] wherein, is a sampling time interval; is a comprehensive abnormality at time t; is a comprehensive abnormality at time t;

[0014] wherein, are change rates of the abnormality in three spatial directions, respectively.

[0015] Preferably, the comparison unit comprises: The comparison unit is configured to compare each key parameter in the measurement state map with a standard range in a second standard map to determine whether the corresponding key parameter exceeds a normal range; Based on statistical characteristics of historical data, the boundary threshold of the standard range is dynamically adjusted

[0016] wherein, , are a mean value and a standard deviation of the historical feature quantity, respectively; is an adjustable coefficient; is a time adjustment function; is a corresponding standard range;

[0017] wherein, , is an adjustment parameter; T is a time period; , are an initial time and a maximum time, respectively.

[0018] ​​​The application provides a vertical furnace body fault diagnosis method, comprising the following steps: Step 1: Real-time monitoring is performed on each working component in the furnace body by using a plurality of measurement devices pre-configured and installed on the vertical furnace body, and a measurement state diagram of each working component at each moment is sequentially constructed in time sequence, wherein the measurement state diagram covers a plurality of key parameter information of the working component and a change trend of each key parameter over time; Step 2: An image acquisition device arranged inside the vertical furnace body is used to acquire a working state diagram inside the furnace body, and image preprocessing is performed on the acquired working state diagram, wherein the preprocessing includes image filtering, noise reduction, contrast enhancement and edge detection; Step 3: The preprocessed working state diagram is compared and analyzed with a first standard diagram of the vertical furnace under a corresponding set running mode, and the difference between the two in image features is calculated, and meanwhile, the second standard diagram of the working component under the set running mode is compared with the constructed measurement state diagram; Step 4: The characteristic quantity of the working component is determined according to the difference between the working state diagram and the first standard diagram and the abnormal comparison result, and the working component is subjected to fault diagnosis to lock the final fault type and give a warning according to a mapping relationship between the characteristic quantity and the fault type pre-established.

[0019] Compared with the prior art, the application has the beneficial effects that: The application constructs a full-closed-loop intelligent system of monitoring-preprocessing-comparison-diagnosis, breaks through the limitations of singleness and staticity of the traditional method, and realizes accurate and early diagnosis of faults.

[0020] Other features and advantages of the application will be described in the following description, and some will become apparent from the description, or will be learned from the practice of the application. The purposes and other advantages of the application can be achieved and obtained by the structures specifically pointed out in the written description and the accompanying drawings.

[0021] The technical solutions of the application will be further described in detail below with reference to the drawings and embodiments. DETAILED DESCRIPTION

[0022] The accompanying drawings are used to provide a further understanding of the application, and constitute a part of the specification, and are used to explain the application together with embodiments of the application, and do not constitute a limitation on the application. In the drawings: Figure 1 It is a structure diagram of a vertical furnace body fault diagnosis system in an embodiment of the application; Figure 2 It is a flowchart of a vertical furnace body fault diagnosis method in an embodiment of the application. DETAILED DESCRIPTION

[0023] The preferred embodiments of the present application are described below in conjunction with the accompanying drawings, in which it is understood that the preferred embodiments described below are merely intended to illustrate and explain the present application, and are not intended to limit the present application.

[0024] The present application provides a vertical furnace body fault diagnosis system, as shown in the accompanying drawings, comprising: Figure 1 A state diagram construction module for real-time monitoring of each working component in the furnace body by using a plurality of measurement devices pre-configured and installed on the vertical furnace body, and sequentially constructing a measurement state diagram of each working component at each time point according to time sequence; An image preprocessing module for acquiring a working state diagram inside the furnace body based on an image acquisition device arranged inside the vertical furnace body, and performing image preprocessing on the acquired working state diagram, the preprocessing including image filtering, noise reduction, contrast enhancement and edge detection; A comparative analysis module for comparing and analyzing each preprocessed working state diagram with a first standard diagram of the vertical furnace under a corresponding set running mode, calculating the difference between the two in image features, and at the same time, performing abnormal comparison according to a second standard diagram of the working component under the set running mode and the constructed measurement state diagram; A fault diagnosis module for determining characteristic quantities of the working component according to the difference between the working state diagram and the first standard diagram and the abnormal comparison result, and locking the final fault type and issuing a warning for the working component according to a pre-established mapping relationship between the characteristic quantities and the fault types. In this embodiment, the plurality of measurement devices include, for example, an infrared thermal imager (for measuring temperature distribution, precision ±0.5℃), a Hall current sensor (for measuring current, precision ±0.1A), and a strain gauge (for measuring thermal expansion deformation, precision ±1μm) for heating element monitoring; a high-precision thermocouple (for measuring its own temperature, precision ±0.1℃) and a MEMS vibration sensor (for measuring vibration, range 0-20g) for thermocouple monitoring.

[0025] In this embodiment, the working components include, for example, heating elements (resistance wire, silicon-carbon rod), thermocouples (K-type and S-type thermocouples), transmission components (gears, belt pulleys), and sealing assemblies (graphite sealing rings, metal bellows).

[0026] In this embodiment, the monitoring logic is classified according to core functions and failure modes, for example, heating elements focus on temperature-current-deformation coupling, and thermocouples focus on temperature deviation-vibration-aging correlation.

[0027] In this embodiment, the monitoring logic is classified according to core functions and failure modes, for example, heating elements focus on temperature-current-deformation coupling, and thermocouples focus on temperature deviation-vibration-aging correlation.

[0028] In this embodiment, the construction method of the state diagram is measured: a visualization module is developed using Python+Matplotlib, and multiple parameters are fused according to the time axis: the horizontal axis is time (resolution 1s), and the vertical axis is divided into three layers: the upper layer displays the temperature curve (red, y-axis range 0-1200℃), the middle layer displays the current curve (blue, y-axis range 0-20A), and the lower layer displays the vibration curve (green, y-axis range 0-5g); the parameter correlation thermogram is superimposed: the Pearson correlation coefficient of temperature and current is calculated to display the coupling relationship with a color gradient (blue→red indicates negative correlation→positive correlation).

[0029] In this embodiment, different components involve different key parameters, such as the key parameters of the heating element: temperature, current, vibration amplitude, deformation, and voltage. Trend analysis: through sliding window fitting (window size 10s, polynomial order 3), the rate of change of the parameter (such as temperature rise rate) is calculated, and sudden rise / sudden drop (rate exceeds 5℃ / s) or continuous drift (deviation exceeds 10℃ and lasts for 30s) is marked.

[0030] In this embodiment, the image acquisition device usually needs to withstand harsh environments such as high temperature, corrosive gas and dust, and in this application, the image acquisition device is a high-temperature-resistant industrial-grade CCD camera (such as Basler acA2040-90um, resolution 2048x2048, frame rate 90fps), which is equipped with a narrow-band filter (to filter out infrared interference in the furnace, with a center wavelength of 650nm), a water cooling system and a protective shell. The high-temperature-resistant industrial-grade CCD camera includes a high-temperature-resistant lens and a CCD sensor. The working temperature of the above image acquisition device can be up to 1800℃; In this embodiment, the working state diagram is an image reflecting the shape, position and temperature distribution of the working components in the furnace, such as the light-emitting profile of the heating element and the spatial posture of the thermocouple; the data format is a 16-bit grayscale image (quantization range 0-65535), stored in TIFF format, with a timestamp and a working condition label (such as “15min in the heating-up phase”).

[0031] In this embodiment, image preprocessing (implemented in stages) is performed: Gaussian filtering: for thermal noise (normal distribution), a 5x5 Gaussian kernel is used to suppress random noise; Median filtering: for dust spots (salt and pepper noise), a 3x3 window is used to replace the center pixel with the median value in the neighborhood.

[0032] Non-local mean (NLM): search for similar blocks (radius 15) and weighted average denoising.

[0033] Contrast enhancement: Contrast Limited Adaptive Histogram Equalization (CLAHE): block (8x8) processing, limit contrast threshold, stretch dark details (e.g. thermocouple edge).

[0034] Edge detection: Canny algorithm: double threshold (low threshold 50, high threshold 150), first calculate gradient amplitude, then extract edge through non-maximum suppression and hysteresis threshold, mark heating element contour (for subsequent positioning).

[0035] First standard image (furnace body level): construction: under the rated operation mode of vertical furnace (e.g. heating power 80kW, holding temperature 850℃, running 2 hours to reach thermal equilibrium), continuously collect 1000 pre-processed working state images, select typical frames (cluster center frames) through K-means clustering, then compress features (retain 95% variance) through principal component analysis (PCA) to generate standard images.

[0036] Feature parameters: gray mean (μ=128), edge density (D=20 edges per 1000 pixels, ), temperature field entropy (H=1.2, reflecting temperature distribution uniformity).

[0037] Image feature difference calculation: gray difference: calculate pixel by pixel, count mean; Edge difference: calculate edge contour difference using Hausdorff distance.

[0038] Second standard image (component level): construction: for a single working component (e.g. heating element), collect 1000 hours of normal operation data, extract parameter distribution of measurement state image (current 10±0.5A, temperature 800±50℃, vibration 0.2±0.1g), establish parameter association model (e.g. linear relationship between current and temperature T=80I-0, residual error) through Bayesian network. Abnormal comparison: calculate parameter deviation and association residual error in real time, double threshold triggers abnormal label.

[0039] Feature quantity is a composite feature that combines image difference and parameter anomaly, such as: Image features: abnormal area, edge distortion rate; Parameter features: current deviation rate, temperature-current association residual error, abnormal duration (length of time exceeding threshold).

[0040] Fusion algorithm: use D-S evidence theory to fuse multiple features.

[0041] Feature quantity-fault type mapping relationship: construction: collect 500+ historical fault data (e.g. heating element short circuit, thermocouple drift, gear wear), use random forest algorithm to train model: input feature vector, output fault type (e.g. "heating element short circuit" probability 0.92, "thermocouple drift" probability 0.05).

[0042] In this embodiment, fault diagnosis and early warning: diagnosis logic: when the confidence Bel (fault) > 0.8, lock the fault, and predict the development stage by combining the fault evolution model (such as short circuit fault current surge -> temperature out of control -> deformation fracture path). Early warning implementation: Minor fault (such as thermocouple drift, confidence 0.8-0.9): system pop-up prompt (yellow), voice warning (frequency 500Hz, duration 2s); Serious fault (such as heating element short circuit, confidence ≥0.9): sound and light alarm (red flashing + 1000Hz buzzer), SMS push (including fault location, disposal suggestion), automatic trigger interlock (cut off heating power, start cooling program).

[0043] The beneficial effects of the above technical scheme are: the present application constructs a full-closed loop intelligent system of monitoring-preprocessing-contrast-diagnosis, breaks through the limitations of single and static of traditional methods, and realizes accurate and early diagnosis of faults.

[0044] The present application provides a vertical furnace body fault diagnosis system, the state diagram construction module comprises: The mode determination unit is used for determining the configuration and installation mode of the measuring device according to the type of the working component and the monitoring requirement, wherein the temperature sensor, the current sensor and the voltage sensor are uniformly arranged on the surface and around the heating element, and the temperature sensor and the vibration sensor are arranged at the measuring end and the connecting end of the thermocouple respectively; The model establishment unit is used for establishing a sensor data space-time calibration model, and eliminating the clock deviation among different sensors based on cross-correlation analysis; The detection and repair unit is used for detecting and repairing the measurement results by adopting 3σ criterion combined with moving average filtering; The correlation analysis unit is used for constructing a parameter correlation matrix, analyzing the nonlinear correlation among parameters through correlation coefficient, eliminating redundant parameters and retaining core characteristic parameters; The visualization unit is used for reducing the dimensionality of the multi-dimensional parameters by using T-SNE algorithm, representing the third dimension parameter value in the two-dimensional coordinate system by color gradient, and forming a three-dimensional measurement state diagram containing parameter real-time value, change trend and correlation, wherein the measurement state diagram covers a variety of key parameter information of the working component and the change trend of each key parameter with time.

[0045] In this embodiment, the type of working component: covers heating elements (such as silicon molybdenum rods, alloy resistance wires), thermocouples (K type / S type thermocouples, used for furnace temperature feedback).

[0046] Monitoring requirements: heating element: monitor thermal distribution uniformity (prevent local overheating and fuse), abnormal electrical parameters (prevent short circuit / open circuit); thermocouple: monitor temperature drift (prevent aging failure), mechanical vibration (prevent loose connection leading to measurement deviation).

[0047] Measurement device selection and deployment: temperature sensor: PT100 platinum resistance is pasted on the surface of the heating element, one every 5 cm; 8 infrared temperature measurement modules (such as MLX90640, field of view angle 110°, covering the whole heating area) are installed in a ring around the heating element. Current / voltage sensor: Hall current sensor (ACS724, range 0-50A, response time 5μs) is connected in series in the power supply loop of the heating element, and differential voltage transmitter (range 0-500V, accuracy ±0.2%) is connected in parallel, which can capture real-time electrical parameter fluctuations. Vibration sensor: MEMS accelerometer (ADXL355, range ±20g, resolution 0.001g) is fixed on the thermocouple connection end (junction box) through a magnetic base, which monitors the vibration amplitude and frequency (prevents signal jump caused by loose connection).

[0048] Space-time calibration model: solves the problem of time asynchrony (clock deviation) and spatial delay (physical response difference caused by different installation positions) of multiple sensors, ensuring consistent data timing.

[0049] Cross-correlation analysis (time calibration): Formula: Calculate the cross-correlation function for temperature sequence T(t) and current sequence I(t).

[0050] Implementation: Search for the peak position of CCF(τ), determine the time deviation, and shift the current data backward by τ* sampling points to achieve clock synchronization accuracy ≤10ms.

[0051] Correlation matrix: N×N matrix, element R(i,j) represents the nonlinear correlation strength of parameters i and j (Spearman rank correlation coefficient is used to adapt to nonlinear relationship).

[0052] In this embodiment, T-SNE algorithm: maps high-dimensional data to 2D plane, preserving local similarity (such as clustering characteristics of electric-thermal coupling).

[0053] Visualization design: two-dimensional coordinates (x,y): represent parameter distribution after dimensionality reduction (perplexity=30, iterate 1000 times, highlight local clustering); color gradient: third dimension (such as temperature T) uses blue→red gradient (200℃→1200℃); dynamic features: point size represents current I (the larger I is, the larger the point diameter is), trajectory line represents time trend (arrow direction is time flow direction). Example: in the heating stage, points move from blue (low temperature) to red (high temperature), trajectory extends to the right and up, points gradually become larger, intuitively presenting the coupling relationship of current rising→temperature rising.

[0054] The beneficial effects of the above technical solution are: through customized sensor deployment to cover multi-physical field monitoring, calibration to eliminate data asynchronous error, 3σ-moving average to purify noise, Spearman coefficient to accurately select core parameters, and finally T-SNE dimension reduction to construct a three-dimensional visual measurement state graph of parameter value+change trend+correlation.

[0055] The present application provides a vertical furnace body fault diagnosis system, further comprising: A graph construction module is configured to, when the vertical furnace body is in a normal running state under a set running mode and each performance index reaches the optimal corresponding set running mode, collect the internal working state graph of the furnace body in multiple time periods and regard it as a first standard graph. A standard determination module is configured to pre-process and extract features of the first standard graph to obtain an image feature vector under the normal running state, and determine the feature parameters and standard range of the first standard graph through statistical analysis and machine learning algorithms.

[0056] In this embodiment, the set running mode is a running state of the vertical furnace preset according to the production process, including heating mode (heating rate 5℃ / min), holding mode (temperature 850±5℃), cooling mode (natural cooling), etc., each mode corresponding to specific control parameters (such as heating power, fan speed), for example, taking a semiconductor wafer annealing furnace as an example, setting a high-temperature annealing mode: target temperature 1000℃, holding time 2 hours, heating power gradient increasing (0→80%→100%), collecting images in the stable stage (1 hour after holding to 30 minutes before holding ends) under this mode.

[0057] The determination conditions of the normal running state are: Parameter stability: key parameters (furnace temperature, heating current, furnace pressure) fluctuation ≤±1% for 30 minutes; No alarm of the equipment: all sensor signals are within the designed threshold (such as thermal couple vibration amplitude <0.5g); Process compliance: wafer resistivity uniformity after annealing ≥98% (verified by offline detection).

[0058] The optimal performance index is the comprehensive optimal parameter of the equipment efficiency, precision, and energy consumption under this running mode, such as: Heating mode: the heating rate reaches the designed maximum value (10℃ / min) and the energy consumption ≤1.2kWh / ℃; Holding mode: temperature uniformity ≤±1.5℃ (temperature difference at different positions in the furnace).

[0059] In this embodiment, the feature extraction is image feature classification: gray scale statistical features, edge features, texture features, and correlation.

[0060] The construction example of the image feature vector is as follows: 10-dimensional feature vectors are extracted from the heating area image.

[0061] The beneficial effects of the above technical solutions are as follows: by collecting images under the optimal working condition of a specific operation mode, combining multi-stage preprocessing (noise suppression, geometric correction) and multi-dimensional feature extraction (gray level, edge, texture), using statistical analysis + PCA dimension reduction to construct a dynamic standard range, the risk of misjudgment caused by working condition changes or outdated standards is significantly reduced, and a reliable digital benchmark is provided for subsequent fault diagnosis.

[0062] The present application provides a vertical furnace body fault diagnosis system, the fault diagnosis module comprises: The comparison unit is used for comparing each key parameter in the measurement state diagram with the standard range in the second standard diagram to determine whether the corresponding key parameter exceeds the normal range; The conformity judgment unit is used for analyzing whether the correlation between the key parameters conforms to the standard correlation specified in the second standard diagram; and comparing the image features in the measurement state diagram with the image features in the second standard diagram to analyze whether the working component has abnormal changes; The feature quantity acquisition unit is used for combining the differences between the working state diagram and the first standard diagram, and using a comprehensive analysis algorithm to fuse and process each comparison result to determine the feature quantity of the abnormal condition of the working component, wherein the feature quantity further includes: the duration of the abnormality of each key parameter, the frequency of the abnormality, and the propagation and correlation of the abnormality between different parameters and image features.

[0063] In this embodiment, the measurement state diagram is a multi-dimensional data diagram containing real-time monitoring parameters (such as the temperature, current, and deformation of the heating element) and time series trends of the working component, and the format is time-stamped structured data (such as CSV, containing columns: time, temperature, current, and vibration amplitude).

[0064] The second standard diagram (component-level standard) is a standard database established based on the design parameters of the working component, historical normal data, and simulation models, and contains: Parameter standard range: such as heating element current Istd=20±1.5A (design range under rated power); Correlation relationship model: such as the temperature difference between the hot junction measurement end and the connection end (normal heat conduction relationship).

[0065] In this embodiment, the parameter correlation relationship is: Standard correlation model: Physical law: T=k0I2R+T0 (Joule's law, k0 is the heat work conversion coefficient, and R is the resistance); Statistical correlation: Spearman correlation coefficient of thermocouple vibration acceleration A and temperature deviation ΔT is ρ=0.85 (the greater the vibration, the greater the deviation).

[0066] Abnormality detection: when the measured T0=1.2kI2R+T0 (residual error is greater than 20℃), it is determined that the thermoelectric coupling is abnormal, which may be caused by the resistance change of the heating element due to oxidation.

[0067] Image feature comparison: second standard image features: The standard edge profile length Lstd of the heating element is 1000±50 pixels (profile without fracture); The standard installation angle θstd of the thermocouple is 90°±2° (perpendicular to the furnace wall, and the angle is detected by Hough line detection).

[0068] Feature quantity: composite feature: Abnormal duration tdur=15min (continuous time of a parameter exceeding the range); Abnormal frequency f=3 times / hour (number of abnormal occurrences per unit time); Propagation path: current abnormality→temperature abnormality after 5min (indicating fault propagation caused by thermal inertia).

[0069] The beneficial effects of the above technical solutions are: through multi-dimensional comparative analysis (parameter value, correlation, image feature) and composite feature fusion, the scheme realizes precise quantization of the working component abnormality.

[0070] The present application provides a vertical furnace body fault diagnosis system, the fault diagnosis module further comprises: A nonlinear mapping unit is configured to collect historical fault data of working components of the vertical furnace, wherein the historical fault data comprises feature quantity data under different fault types, and a nonlinear mapping model between the feature quantity data and the fault types is established by training and learning the feature quantity data using a machine learning algorithm; A fault locking unit is configured to input the determined feature quantity into the established mapping model during the fault diagnosis process, output possible fault types and probabilities of the possible fault types, and lock a final fault type according to a preset probability threshold and a fault diagnosis rule; A mode formulating unit is configured to formulate different warning levels and warning modes according to the severity of the fault type and possible influences.

[0071] Preferably, the warning levels and warning modes comprise: When it is a slight fault, a low-intensity sound warning and a system interface prompt warning are adopted; When it is a serious fault, high-intensity sound warning, flashing light warning, and sending short message and email warning to relevant personnel, and automatically triggering the corresponding emergency handling measures are adopted.

[0072] In this embodiment, the data composition of historical fault data is as follows: Fault type: such as heating element short circuit (type F1), thermocouple aging (F2), and transmission component jamming (F3); characteristic quantity: including parameter abnormality degree (such as current deviation rate I=15%), image feature (such as heating wire profile shortening by 10%), and time attribute (abnormality lasting for 15 min).

[0073] Data collection is extracted from 3 sets of same type vertical furnaces 5 years of operation records, and a total of 200 fault samples (F1: 80, F2: 60, and F3: 60) are obtained.

[0074] In this embodiment, the nonlinear mapping model is as follows: , wherein, is the output of the jth type of neural network (such as j=1 corresponding to F1 fault probability).

[0075] In this embodiment, the setting rule of the probability threshold is as follows: Single high probability: when P(Fj)>0.8, the fault type Fj is directly locked; Multiple fault competition: when P(Fj)∈[0.4,0.8] and |P(Fj)-P(Fk)|<0.2, a compound fault judgment is triggered (such as F1 and F2 exist at the same time).

[0076] In this embodiment, the fault diagnosis rule is as follows: Physical constraint: if the current abnormally rises but the temperature does not synchronously rise (violating Joule's law), F1 (short circuit leading to resistance reduction) is preferentially determined instead of F2 (thermocouple fault leading to temperature measurement abnormality); Time sequence rule: if vibration abnormality (F3 precursor) appears prior to temperature abnormality (F1 / F2), a secondary problem caused by mechanical fault is preferentially considered.

[0077] In this embodiment, the severity classification is as follows: Classification standard table

[0078] Warning mode: classified response: F1 (emergency): Acoustic and light alarm: red LED stroboscopic (frequency 2 Hz) + 1000 Hz buzzer; Automatic control: trigger safety interlock (cut off heating power and start inert gas protection); Notification chain: push SMS to operation manager within 10 seconds (including fault code, positioning map), and send historical similar case processing scheme by email attachment.

[0079] F2 (important): Interface warning: a yellow warning box pops up on the HMI interface, marking abnormal parameters; Voice prompt: broadcast hot couple system abnormal, please check calibration (Mandarin + English bilingual); Work order generation: automatically create a preventive maintenance work order and assign it to the on-duty technician.

[0080] The beneficial effects of the above technical solutions are: through historical data driven modeling, probability decision mechanism and hierarchical warning system, the intelligentization and precision of fault diagnosis are realized.

[0081] The present application provides a vertical furnace body fault diagnosis system, further comprising: The self-learning module is used for inputting new fault data and diagnosis results into the self-learning module when a new fault type is diagnosed or it is found that the existing mapping relationship has errors, and adjusting and optimizing the first standard graph, the second standard graph and the mapping relationship through a machine learning algorithm.

[0082] The beneficial effects of the above technical solutions are: through adjustment and optimization, the accuracy of the algorithm is ensured, and an accurate basis is provided for diagnosis.

[0083] The present application provides a vertical furnace body fault diagnosis system, and the characteristic quantity acquisition unit comprises: The first extraction sub-block is used for extracting the difference feature vector of the working state graph and the first standard graph in the gray space Wherein, The i-th difference feature is represented as n, which represents the total number of difference features; The second extraction sub-block is used for extracting the difference feature vector of the temperature field distribution Wherein, The temperature deviation of the j-th temperature sampling point is represented as m, which represents the total number of temperature sampling points; The third extraction sub-block is used for extracting the difference feature vector of the texture feature Wherein, The difference percentage of the u-th texture feature is represented as k, which represents the number of texture features; The construction sub-block is used for constructing a time sequence weight matrix for the difference feature vector of the continuous N time points; The distribution calculation sub-block is used for constructing a dynamic Bayesian fusion model based on the Markov chain Monte Carlo method, combining the difference feature vector of the N time points and the time sequence weight matrix, and calculating the posterior probability distribution through the Gibbs sampling algorithm; The feature fusion sub-block is used for constructing a fuzzy neural network based on a posterior probability distribution for feature fusion, wherein a three-layer fuzzy neural network is designed, the difference feature vector is used as the input layer, the Gaussian membership function is used in the hidden layer, and the weighted summation function is used in the output layer; The feature quantity determination sub-block is used for calculating the comprehensive abnormality index, the time evolution trend of the feature quantity and the spatial distribution gradient of the feature quantity according to the feature fusion result, and determining the feature quantity of the abnormal condition of the working component operation state;

[0084] Among them, is the sampling time interval; is the comprehensive abnormality at time t; is the comprehensive abnormality at time t; is the comprehensive abnormality at time t;

[0085] Among them, are the change rates of the abnormality in three spatial directions, respectively.

[0086] In this embodiment, the gray space: the numerical expression of the brightness of the image pixels (such as 8-bit image 0-255), reflecting the light and dark distribution of the working component (such as the light intensity difference of the heating element). Temperature field distribution: three-dimensional temperature grid (spatial resolution 1cm) constructed by infrared thermal imager (such as FLIRA655sc) or multi-sensor interpolation.

[0087] Texture feature: contrast, entropy, correlation (reflecting surface roughness, uniformity) are extracted by using gray level co-occurrence matrix (GLCM).

[0088] The beneficial effects of the above technical solutions are: through the technical chain of multi-domain features, spatio-temporal modeling and probability fusion, the core pain points of one-sided features, rigid threshold and poor dynamic response in the fault diagnosis of the vertical furnace body are solved, and the leap from passive alarm to active prediction and accurate positioning is realized, which provides a quantifiable and extensible technical paradigm for intelligent operation and maintenance of high-end equipment.

[0089] The present application provides a vertical furnace body fault diagnosis system, and the comparison unit comprises: The comparison unit is used for comparing each key parameter in the measurement state diagram with the standard range in the second standard diagram to determine whether the corresponding key parameter exceeds the normal range; Based on the statistical characteristics of historical data, the boundary threshold of the standard range is dynamically adjusted ;

[0090] Among them,​ , are respectively the mean value and the standard deviation of the historical characteristic quantity; is an adjustable coefficient; is a time adjustment function; is the corresponding standard range;

[0091] wherein, , is an adjustment parameter; T is a time period; , are respectively the initial time and the maximum time.

[0092] The beneficial effects of the above technical solutions are: based on the periodic term adapt to the diurnal temperature difference, based on the aging term adapt to the device aging, realize the reasonable adjustment of the threshold value.

[0093] The present application provides a vertical furnace body fault diagnosis method, as shown in the figure, comprising the following steps: Figure 2 Step 1: using a variety of measurement devices installed on the vertical furnace body in advance, real-time monitoring of each working component in the furnace body, sequentially constructing the measurement state diagram of each working component at each time according to time sequence, the measurement state diagram covers a variety of key parameter information of the working component and the change trend of each key parameter with time; Step 2: based on the image acquisition device arranged inside the vertical furnace body, collecting the working state diagram inside the furnace body, and carrying out image preprocessing on the collected working state diagram, the preprocessing includes: image filtering, noise reduction, contrast enhancement and edge detection; Step 3: comparing and analyzing each preprocessed working state diagram with the first standard diagram of the vertical furnace under the corresponding set running mode, calculating the difference between the two in image features, and at the same time, comparing the second standard diagram of the working component under the set running mode with the constructed measurement state diagram; Step 4: determining the characteristic quantity of the working component according to the difference between the working state diagram and the first standard diagram and the abnormal comparison result, and locking the final fault type and early warning of the working component according to the pre-established mapping relationship between the characteristic quantity and the fault type. The beneficial effects of the above technical solutions are: the present application constructs a full-closed-loop intelligent system of monitoring-preprocessing-comparison-diagnosis, breaks through the single and static limitations of traditional methods, and realizes the accurate and early diagnosis of faults.

[0094]

[0095] ​Obviously, many modifications and variations of the present application are possible in light of the above teachings. It is, therefore, to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A vertical furnace furnace body failure diagnosis system characterized by, The method comprises the following steps: A state diagram construction module is configured to utilize a plurality of measurement devices pre-installed on the furnace body of the vertical furnace to monitor each working component in the furnace body in real time, and to sequentially construct a measurement state diagram of each working component at each time point according to time sequence; An image preprocessing module is configured to acquire a working state diagram inside the furnace body based on an image acquisition device arranged inside the furnace body of the vertical furnace, and to perform image preprocessing on the acquired working state diagram, wherein the preprocessing includes image filtering, noise reduction, contrast enhancement, and edge detection; A comparative analysis module is configured to compare and analyze each preprocessed working state diagram with a first standard diagram of the vertical furnace under a corresponding set running mode, to calculate the difference between the two in image features, and to simultaneously perform abnormal comparison according to a second standard diagram of the working component under the set running mode and the constructed measurement state diagram; A fault diagnosis module is configured to determine characteristic quantities of the working component according to the difference between the working state diagram and the first standard diagram and the abnormal comparison result, and to perform fault diagnosis on the working component to lock the final fault type and issue a warning according to a pre-established mapping relationship between the characteristic quantities and the fault types.

2. The vertical furnace body failure diagnosis system according to claim 1, characterized by The state diagram construction module comprises: A mode determination unit is configured to determine the configuration and installation mode of the measurement devices according to the type and monitoring requirements of the working component, wherein temperature sensors, current sensors, and voltage sensors are uniformly arranged on the surface and around the heating element, and temperature sensors and vibration sensors are arranged at the measurement end and the connection end of the thermocouple, respectively; A model establishment unit is configured to establish a sensor data space-time calibration model, and to eliminate the clock bias between different sensors based on cross-correlation analysis; A detection and repair unit is configured to detect and repair abnormal values of the measurement results by adopting the 3σ criterion combined with moving average filtering; A correlation analysis unit is configured to construct a parameter correlation matrix, to analyze the nonlinear correlation between parameters through correlation coefficients, to eliminate redundant parameters, and to retain core characteristic parameters; A visualization unit is configured to perform dimension reduction visualization on multi-dimensional parameters by using the T-SNE algorithm, to represent the third-dimensional parameter value in a two-dimensional coordinate system by using a color gradient, and to form a three-dimensional measurement state diagram containing real-time parameter values, change trends, and correlation relationships, wherein the measurement state diagram covers a plurality of key parameter information of the working component and the change trend of each key parameter over time.

3. The vertical furnace body failure diagnosis system according to claim 1, characterized by Further comprising: A diagram construction module is configured to acquire working state diagrams inside the furnace body in a plurality of time periods when the vertical furnace is in a normal running state under a set running mode and each performance indicator reaches the optimum corresponding to the set running mode, and to regard the working state diagrams as first standard diagrams; A standard determination module is configured to perform preprocessing and feature extraction on the first standard diagrams to obtain image feature vectors under the normal running state, and to determine characteristic parameters and standard ranges of the first standard diagrams by statistical analysis and machine learning algorithms.

4. The vertical furnace body failure diagnosis system according to claim 1, characterized by The fault diagnosis module comprises: A comparison unit is configured to compare each key parameter in the measurement state diagram with the standard range in the second standard diagram, and to determine whether the corresponding key parameter exceeds the normal range. The judgment unit is configured to analyze whether the correlation between the key parameters meets the standard correlation in the second standard graph, and compare the image features in the measurement state graph with the image features in the second standard graph to analyze whether the working component has an abnormal change. The feature quantity acquisition unit is configured to combine the difference between the working state graph and the first standard graph, and fuse the comparison results by using a comprehensive analysis algorithm to determine the feature quantity of the abnormal working component state, wherein the feature quantity further includes: duration of abnormality of each key parameter, frequency of abnormality, and propagation and correlation of abnormality between different parameters and image features.

5. The vertical furnace body failure diagnosis system according to claim 1, characterized by The fault diagnosis module further includes: The nonlinear mapping unit is configured to collect historical fault data of the working component of the vertical furnace, wherein the historical fault data includes feature quantity data under different fault types, and a nonlinear mapping model between the feature quantity data and the fault types is established by using a machine learning algorithm to train and learn the feature quantity data; The fault locking unit is configured to input the determined feature quantity into the established mapping model during the fault diagnosis process, output a possible fault type and a probability of the possible fault type, and lock a final fault type according to a preset probability threshold and a fault diagnosis rule. The mode making unit is configured to make different warning levels and warning modes according to the severity of the fault type and the possible influence.

6. The vertical furnace body failure diagnosis system according to claim 5, characterized by The warning levels and warning modes include: When it is a slight fault, a low-intensity sound warning and a system interface prompt warning are used; When it is a serious fault, a high-intensity sound warning, a flashing light warning, a short message warning and an email warning are sent to relevant personnel, and corresponding emergency handling measures are automatically triggered.

7. The vertical furnace body failure diagnosis system according to claim 1, characterized by Further including: The self-learning module is configured to input new fault data and diagnosis results into the self-learning module when a new fault type is diagnosed or an error exists in the existing mapping relationship, and adjust and optimize the first standard graph, the second standard graph and the mapping relationship by using a machine learning algorithm.

8. The vertical furnace body failure diagnosis method according to claim 4, characterized by, The feature quantity acquisition unit includes: The first extraction sub-block is configured to extract a difference feature vector of the working state graph and the first standard graph in the gray space wherein, represents the i-th difference feature; n represents the total number of difference features; A second extraction sub-block is configured to extract a difference feature vector of the temperature field distribution wherein, represents the temperature deviation of the jth temperature sampling point; m represents the total number of temperature sampling points; a third extraction sub-block for extracting a difference feature vector of the texture feature wherein, denotes the difference percentage of the u-th texture feature; k denotes the number of texture features; The construction sub-block is configured to construct a time sequence weight matrix for the difference feature vectors of the continuous N time points; The distribution calculation sub-block is configured to construct a dynamic Bayesian fusion model based on a Markov chain Monte Carlo method, and combine the difference feature vectors of the N time points and the time sequence weight matrix to calculate a posterior probability distribution by using a Gibbs sampling algorithm; The feature fusion sub-block is configured to construct a fuzzy neural network based on the posterior probability distribution for feature fusion, wherein a three-layer fuzzy neural network is designed, the input layer is the difference feature vector, the hidden layer uses a Gaussian membership function, and the output layer uses a weighted summation function; The feature quantity determination sub-block is used to calculate the comprehensive anomaly index and the time evolution trend of the feature quantity based on the feature fusion results. And the spatial distribution gradient of the feature quantities, and determine the feature quantities of abnormal operating conditions of the working components; wherein is a sampling time interval; is an integrated abnormality degree at time t; is an integrated abnormality degree at time t. wherein respectively, are the rates of change of the abnormality degree in the three spatial directions.

9. The vertical furnace body failure diagnosis method according to claim 4, characterized by, The comparison unit includes: The comparison unit is configured to compare the key parameters in the measurement state graph with the standard range in the second standard graph to determine whether the corresponding key parameter is out of the normal range. Dynamic adjustment of the boundary threshold of the standard range based on statistical properties of historical data ; wherein, , are the mean and standard deviation of the historical feature value, respectively; is an adjustable coefficient; is a time adjustment function; is the corresponding standard range; wherein, , are tuning parameters; T is a time period; , are initial and maximum times, respectively.

10. A method for diagnosing a fault of a shaft furnace body, characterized by, The method includes the following steps: Step 1: Real-time monitoring of each working component in the furnace body is performed by using a plurality of measurement devices pre-installed on the furnace body of the vertical furnace, and a measurement state diagram of each working component at each time is sequentially constructed in time sequence, the measurement state diagram covering a plurality of key parameter information of the working component and a change trend of each key parameter over time; Step 2: An image acquisition device arranged inside the furnace body of the vertical furnace is used to acquire a working state diagram of the inside of the furnace body, and the acquired working state diagram is pre-processed, the pre-processing including image filtering, noise reduction, contrast enhancement and edge detection; Step 3: The pre-processed working state diagram is compared and analyzed with a first standard diagram of the vertical furnace under a corresponding set running mode, a difference between the two in image features is calculated, and at the same time, an abnormal comparison is performed between a second standard diagram of the working component under the set running mode and the constructed measurement state diagram; Step 4: Feature quantities of the working component are determined according to the difference between the working state diagram and the first standard diagram and the abnormal comparison result, and a final fault type and a warning are locked by performing fault diagnosis on the working component according to a mapping relationship between the feature quantities and the fault types pre-established.

Images