Roasting workshop safety risk assessment system for carbon anode production

Abstract

The invention discloses a roasting workshop safety risk assessment system for carbon anode production, particularly relates to the field of data analysis, and comprises a multi-dimensional data entry module, a preprocessing module, an index calculation module and a safety risk assessment module. According to the invention, a process-equipment-environment-material multi-dimensional data acquisition system is constructed, a special sensor array and a slide rail type robot are combined to realize total-factor monitoring and eliminate a space blind area, preprocessing technologies such as wavelet denoising and dynamic standardization are innovatively adopted, the parameter sensitivity is enhanced, and the data quality is improved; based on four-dimensional state space modeling and a dynamic mahalanobis distance algorithm, multi-factor coupling risks are quantified, evaluation timeliness is ensured in combination with a quarterly updated sliding window standard deviation mechanism, the composite risks of process disturbance, equipment decline, environment abnormity and material deterioration can be accurately recognized, blue, yellow, orange and red four-level dynamic early warning is achieved, and the method is suitable for large-scale popularization and application. And multi-dimensional safety decision support covering the whole process is provided for the roasting workshop.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and more particularly to a safety risk assessment system for a baking workshop used in carbon anode production. Background Art

[0002] Carbon anode production is a vital part of the electrolytic aluminum industry. Its roasting workshop involves high-temperature processes, complex equipment, and hazardous substances. Safety risk management is directly related to production safety and efficiency. Traditional safety management mainly relies on manual inspections and single-point monitoring, making it difficult to achieve real-time dynamic risk assessment.

[0003] Existing safety assessment systems typically use single-dimensional monitoring data, such as temperature or gas concentration, to make risk assessments through threshold alarm mechanisms. Data collection is often limited to some key equipment, using simple moving average or fixed threshold processing. Assessment models are mostly linear weighted calculations, and risk level classification relies on static standards.

[0004] However, such methods suffer from problems such as the lack of multi-dimensional data coupling analysis, insufficient capture of the dynamic evolution of environmental risks, and delayed monitoring of material degradation processes. Specifically, they are unable to effectively identify the correlation effects between process disturbances and equipment failures, have difficulty quantifying the spatial heterogeneity of environmental factors, and lack the ability to conduct real-time assessments of internal defects in anode materials, resulting in insufficient early warning accuracy and timeliness. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides a safety risk assessment system for a baking workshop for carbon anode production. Through the following scheme, it solves the problems proposed in the above-mentioned background technology, such as the lack of multi-dimensional data coupling analysis, insufficient capture of the dynamic evolution of environmental risks, and delayed monitoring of material degradation, which lead to low early warning accuracy, poor timeliness, and inability to assess complex risks.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a safety risk assessment system for a baking workshop for carbon anode production, comprising:

[0007] Multi-dimensional data entry module: used to collect process disturbance monitoring data, equipment health status data, environmental risk coupling data, and material property derivative data of the target roasting workshop, and transmit the collected data to the preprocessing module;

[0008] Preprocessing module: used to preprocess the data collected by the multidimensional data entry module and transmit the preprocessed data to the indicator calculation module;

[0009] Index calculation module: used to establish a mathematical model to calculate indicators of the data preprocessed by the preprocessing module, including the process disturbance index calculation model, equipment health calculation model, environmental risk value calculation model and material degradation index calculation model, and transmit the calculation results to the safety risk assessment module;

[0010] Security risk assessment module: Based on the calculation results of the indicator calculation module, a multi-level fusion assessment architecture is used to quantify risks.

[0011] Preferably, the process disturbance monitoring data include the standard deviation of the temperature gradient of carbon blocks in the roasting furnace, the asphalt flue gas escape rate, the filler osmotic pressure fluctuation frequency and the cooling water system temperature difference attenuation coefficient; the equipment health status data include the annular furnace drive motor current harmonic distortion rate, the hydraulic push rod displacement lag time, the burner nozzle carbon deposition vibration amplitude and the refractory material micro-strain rate; the environmental risk coupling data include the spatial variation coefficient of the CO concentration in the dead corner of the workshop, the ground electrostatic potential gradient of the asphalt pump room, the flue gas purification tower pressure difference fluctuation entropy and the instantaneous wind speed vector anomaly in the loading and unloading area; the material property derivative data include the cumulative energy of acoustic emission of cracks in the anode blank, the infrared reflectivity of the residual sodium penetration depth, the dielectric loss angle of the binder coking process and the Lab value of the oxidation chromaticity of the anode surface out of the furnace.

[0012] Preferably, the standard deviation of the temperature gradient of the carbon blocks in the roasting furnace is obtained by real-time scanning of a multi-zone infrared thermal imager array installed on the top of the furnace, and the thermal imager captures the temperature distribution of four predefined observation surfaces at a frequency of 5 frames per second; the asphalt flue gas escape rate is obtained by a workshop ceiling sliding rail gas sampling robot equipped with an online mass spectrometer, which moves along the process axis for sampling every 15 minutes, and the deviation value is calculated by comparing the real-time concentration with the theoretical volatilization model of the DCS system; the filler osmotic pressure fluctuation frequency is monitored by a 16-channel piezoresistive sensor array pre-buried at different depths in the filling layer, and the data is counted by the edge computing gateway for the number of pulses exceeding the ±5kPa threshold per hour; the temperature difference attenuation coefficient of the cooling water system is collected in real time by a PT100 temperature sensor group installed on the water inlet main and the water outlet branch, and the heat exchange efficiency decline trend within a 30-minute window is calculated by the temperature difference change rate algorithm.

[0013] Preferably, the harmonic distortion rate of the annular furnace drive motor current is collected by a clamp-on current sensor in combination with a portable power quality recorder, and the proportion of 1-25 harmonic components is continuously monitored and calculated at the power input end of the motor control cabinet; the hydraulic push rod displacement lag time is synchronized with the PLC control system clock using a laser displacement sensor, and is calculated by comparing the time difference between the rising edge of the valve island control signal and the actual displacement of the push rod reaching the set value; the carbon deposition vibration amplitude of the burner nozzle is measured using a magnetic piezoelectric acceleration sensor installed on the nozzle base, and the proportion of vibration energy in the 5-8kHz frequency band is extracted by fast Fourier transform; the micro-strain rate of the refractory material is monitored in real time by a fiber optic Bragg grating sensor network embedded in the refractory layer of the furnace wall, and 3 measuring points are arranged per meter to measure the wavelength drift caused by thermal expansion.

[0014] Preferably, the spatial variation coefficient of CO concentration in dead corners of the workshop is formed by a detection grid composed of 32 electrochemical gas sensors deployed in a distributed manner. After synchronous data collection every 5 minutes, the ratio of the standard deviation of the concentration in each area to the average value is calculated; the electrostatic potential gradient of the ground in the asphalt pump room is measured by a rotating electrostatic field strength meter installed on the chassis of the inspection robot, and the ground potential value is recorded every 0.5 meters along the material conveying path to generate a gradient curve; the pressure difference fluctuation entropy value of the flue gas purification tower is measured by installing a high-precision differential pressure transmitter at the inlet and outlet of the tower body, and the raw data is acquired at a sampling frequency of 100Hz and the permutation entropy value within a 10-minute window is calculated; the instantaneous wind speed vector anomaly in the loading and unloading area is monitored by a monitoring array composed of 4 groups of three-dimensional ultrasonic anemometers, and the anomaly index is calculated by comparing the real-time wind speed direction with the Mahalanobis distance of the historical working condition database.

[0015] Preferably, the cumulative acoustic emission energy of the internal cracks of the anode blank is attached to the mold surface using a 32-channel acoustic emission sensor network, and the acoustic emission signals in the 30-150kHz frequency band are collected in real time during the roasting process and the energy cumulative value is calculated; the infrared reflectivity of the sodium penetration depth of the residual anode is scanned by an online near-infrared spectrometer installed on the cooling conveyor belt to extract the reflection intensity ratio of the 1350nm characteristic band; the dielectric loss angle of the binder coking process is embedded in the molding mold using a microwave resonant cavity sensor to monitor the change of the dielectric loss factor at a frequency of 2.45GHz in real time; the Lab value of the oxidation chromaticity of the surface of the baked anode is captured by a machine vision system integrated with a high-resolution industrial camera, equipped with a D65 standard light source, and the chromaticity coordinate value is output in real time using the CIE chromaticity space conversion algorithm.

[0016] Preferably, the preprocessing module normalizes the standard deviation of the roasting furnace temperature gradient standard deviation data using a sliding window standard deviation with a window width of 30 seconds, and uses wavelet threshold denoising to eliminate optical interference from the thermal imager; the asphalt smoke emission rate data is smoothed by a moving average filter and then dynamically differentiated with the DCS theoretical value; the number of filler osmotic pressure pulses is subjected to a Poisson distribution test to eliminate random noise and converted into an event rate per unit time; the cooling water temperature difference attenuation coefficient uses the first-order derivative to calculate the temperature difference change slope, and then the seasonal temperature influence is eliminated by Z-score standardization.

[0017] Preferably, the pre-processing module extracts the 3rd / 5th / 7th harmonic energy proportion after FFT spectrum correction of the current harmonic distortion rate, and uses the Kaiser window function to suppress spectrum leakage; the hydraulic push rod delay data is compensated by PLC clock synchronization, and quartiles are implemented to remove abnormal values during non-working periods; the vibration frequency band energy proportion is enhanced by using the Mel frequency cepstral coefficient improved algorithm; the optical fiber strain data is eliminated by the temperature compensation formula to eliminate thermal expansion pseudo-strain.

[0018] Preferably, the preprocessing module uses Kriging spatial interpolation to fill the sensor blind area for the CO concentration variation coefficient, and improves data normality through Box-Cox transformation; the electrostatic gradient curve is smoothed by Savitzky-Golay differential method to extract the curvature extreme points; the pressure difference entropy value is calculated using an improved permutation entropy algorithm and embedded in the Tau automatic delay selection mechanism; the wind speed anomaly is compressed into a three-dimensional vector through principal component analysis to construct a Mahalanobis distance control chart.

[0019] Preferably, the preprocessing module performs wavelet packet decomposition on the acoustic emission energy to extract the energy integral of the crack sensitive frequency band; the near-infrared reflectivity adopts SNV standard normal variable transformation to eliminate surface scattering interference; the dielectric loss angle data is fitted by Debye model to separate the temperature-frequency cross-influence; the chromaticity Lab value uses the CIEDE2000 color difference formula to construct the oxidation degree index.

[0020] Preferably, the process disturbance index calculation model is specifically expressed as:

[0021]

[0022] ΔT'_std: normalized standard deviation of temperature gradient, reflecting the uniformity of heat distribution in the furnace;

[0023] Q_dev: The deviation rate between the measured and theoretical values of asphalt flue gas concentration, which represents the abnormal degree of volatilization control;

[0024] λ_pulse: The number of pulses exceeding the osmotic pressure limit per unit time, indicating the stability of the filler;

[0025] α_cool: Normalized cooling efficiency attenuation coefficient, reflecting the degradation state of the heat exchange system.

[0026] Preferably, the device health calculation model is specifically expressed as:

[0027]

[0028] x1 = THD_m: main harmonic distortion rate, k1 = 15%, w1 = 0.4;

[0029] x2 = τ_hyd: hydraulic delay time, k2 = 120ms, w2 = 0.3;

[0030] x3 = E_vib: vibration energy ratio, k3 = 8%, w3 = 0.2;

[0031] x4=ε_real: true material strain, k4=500μm / m, w4=0.1.

[0032] Preferably, the environmental risk value calculation model is specifically expressed as:

[0033]

[0034] CV_co: coefficient of variation of CO concentration, the larger the value, the more uneven the distribution;

[0035] κ_es: extreme value of electrostatic gradient curvature, reflecting the degree of charge accumulation;

[0036] H_p: pressure difference permutation entropy, representing the degree of chaos in the system;

[0037] D_wind: wind speed Mahalanobis distance, the degree to which abnormal airflow deviates from normal operating conditions.

[0038] Preferably, the material degradation index calculation model is specifically expressed as:

[0039]

[0040] E_ae: crack acoustic emission energy, the larger the value, the more serious the internal defect;

[0041] R_Na: Sodium penetration reflectivity ratio, the smaller the value, the greater the penetration depth;

[0042] tanδ_e: Equivalent dielectric loss angle, reflecting the degree of incomplete coking of the binder;

[0043] ΔE_ox: Oxidation color difference, a large value indicates severe surface oxidation.

[0044] Preferably, the specific process of the security risk assessment module includes:

[0045] S1: Establish a four-dimensional state space:

[0046] is a four-dimensional eigenvector;

[0047] S2: Calculate the degree of deviation between the current state and the historical safety benchmark through Mahalanobis distance: D risk is the risk deviation, is the historical security benchmark vector, T is the transposed vector;

[0048] S3: Risk level classification: Risk levels are divided into L1, L2, L3 and L4, L1 is a blue warning, risk <2σ, trigger, L2 is yellow warning, when 2σ≤D risk <3σ, trigger, L3 is orange warning, when 3σ≤D risk L4 is a red alert and is triggered when D is less than 4σ. L4 is a red alert and is triggered when D ≥ 4σ. σ represents the dynamically updated standard deviation, which is updated every quarter. The sliding window standard deviation is used, and the window = 90 days.

[0049] The technical effects and advantages of the present invention are as follows:

[0050] 1. This invention builds a multi-dimensional data acquisition system that integrates four types of parameters: process, equipment, environment, and materials. It uses a dedicated sensor array and intelligent algorithms to achieve real-time monitoring of all factors. It uses infrared thermal imagers, fiber grating sensors, and other equipment to capture spatial distribution characteristics. It also combines a slide-type sampling robot to eliminate monitoring blind spots, significantly improving data dimensionality and acquisition accuracy.

[0051] 2. This invention innovatively introduces a multi-level preprocessing mechanism, implementing wavelet denoising, spatial interpolation, and dynamic normalization for different data types. It optimizes feature extraction through an improved permutation entropy algorithm and principal component analysis, effectively eliminating noise interference and enhancing the sensitivity of key parameters, providing a high-quality data foundation for subsequent indicator calculations.

[0052] 3. The present invention adopts four-dimensional state space modeling and dynamic Mahalanobis distance evaluation architecture to realize the coupled analysis of multiple risk factors. Through the quarterly updated sliding window standard deviation mechanism, the risk assessment benchmark remains timely. The graded early warning system can accurately identify the combined risks of process disturbances, equipment degradation, environmental anomalies and material degradation, providing precise decision-making support for workshop safety management. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a schematic diagram of the overall structure of the present invention. DETAILED DESCRIPTION

[0054] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0055] refer to Figure 1 A safety risk assessment system for a baking shop for carbon anode production is shown, comprising:

[0056] Multi-dimensional data entry module: used to collect process disturbance monitoring data, equipment health status data, environmental risk coupling data and material property derivative data of the target roasting workshop, and transmit the collected data to the preprocessing module.

[0057] The process disturbance monitoring data include the standard deviation of the temperature gradient of carbon blocks in the roasting furnace, the asphalt flue gas escape rate, the filler osmotic pressure fluctuation frequency and the cooling water system temperature difference attenuation coefficient; the equipment health status data include the harmonic distortion rate of the ring furnace drive motor current, the hydraulic push rod displacement lag time, the burner nozzle carbon deposition vibration amplitude and the refractory material micro-strain rate; the environmental risk coupling data include the spatial variation coefficient of the CO concentration in the dead corner of the workshop, the electrostatic potential gradient on the ground of the asphalt pump room, the pressure difference fluctuation entropy value of the flue gas purification tower and the instantaneous wind speed vector anomaly in the loading and unloading area; the material property derivative data include the cumulative energy of the acoustic emission of cracks in the anode blank, the infrared reflectivity of the residual sodium penetration depth, the dielectric loss angle of the binder coking process and the Lab value of the oxidation chromaticity of the anode surface out of the furnace.

[0058] The standard deviation of the temperature gradient of the carbon blocks in the roasting furnace is obtained by real-time scanning of a multi-area infrared thermal imager array installed on the furnace top. The thermal imager captures the temperature distribution of four predefined observation surfaces at a frequency of 5 frames per second; the asphalt flue gas escape rate is obtained by a workshop ceiling sliding rail gas sampling robot equipped with an online mass spectrometer, which moves along the process axis for sampling every 15 minutes, and the deviation value is calculated by comparing the real-time concentration with the theoretical volatilization model of the DCS system; the frequency of the filler osmotic pressure fluctuation is monitored by a 16-channel piezoresistive sensor array pre-buried at different depths in the filler layer, and the data is counted by the edge computing gateway for the number of pulses exceeding the ±5kPa threshold per hour; the temperature difference attenuation coefficient of the cooling water system is collected in real time by a PT100 temperature sensor group installed on the water inlet main and the water outlet branch, and the trend of heat exchange efficiency decline within a 30-minute window is calculated by the temperature difference change rate algorithm.

[0059] The harmonic distortion rate of the annular furnace drive motor current is collected through a clamp-on current sensor in conjunction with a portable power quality recorder, and the proportion of 1-25 harmonic components is continuously monitored and calculated at the power input end of the motor control cabinet; the hydraulic push rod displacement lag time is synchronized with the PLC control system clock using a laser displacement sensor, and is calculated by comparing the time difference between the rising edge of the valve island control signal and the actual displacement of the push rod reaching the set value; the carbon deposit vibration amplitude of the burner nozzle is measured using a magnetic piezoelectric acceleration sensor installed on the nozzle base, and the proportion of vibration energy in the 5-8kHz frequency band is extracted through fast Fourier transform; the micro-strain rate of the refractory material is monitored in real time by a fiber optic Bragg grating sensor network embedded in the refractory layer of the furnace wall, and three measuring points are arranged per meter to measure the wavelength drift caused by thermal expansion.

[0060] The spatial variation coefficient of CO concentration in dead corners of the workshop is measured using a distributed detection grid composed of 32 electrochemical gas sensors. Data is collected synchronously every 5 minutes, and the ratio of the standard deviation of the concentration in each area to the mean is calculated. The electrostatic potential gradient of the ground in the asphalt pump room is measured using a rotating electrostatic field strength meter installed on the chassis of the inspection robot. The ground potential value is recorded every 0.5 meters along the material conveying path to generate a gradient curve. The flue gas purification tower pressure difference fluctuation entropy value is measured by installing high-precision differential pressure transmitters at the tower inlet and outlet. Raw data is acquired at a sampling frequency of 100 Hz, and the permutation entropy value within a 10-minute window is calculated. The instantaneous wind speed vector anomaly in the loading and unloading area is monitored using a monitoring array consisting of four groups of three-dimensional ultrasonic anemometers. The anomaly index is calculated by comparing the real-time wind speed direction with the Mahalanobis distance in the historical operating condition database.

[0061] The cumulative acoustic emission energy of cracks inside the anode blank is measured using a 32-channel acoustic emission sensor network attached to the mold surface. During the roasting process, acoustic emission signals in the 30-150 kHz frequency band are collected in real time and the energy cumulative value is calculated. The infrared reflectivity of the sodium penetration depth of the residual anode is scanned by an online near-infrared spectrometer installed on the cooling conveyor belt to scan the surface of the residual anode and extract the reflection intensity ratio of the 1350 nm characteristic band. The dielectric loss angle during the coking process of the binder is measured using a microwave resonant cavity sensor embedded in the molding mold to monitor the change of the dielectric loss factor at a frequency of 2.45 GHz in real time. The Lab value of the oxidation chromaticity of the surface of the baked anode is captured by a machine vision system integrated with a high-resolution industrial camera, equipped with a D65 standard light source, and using a CIE color space conversion algorithm to output the chromaticity coordinate value in real time.

[0062] Preprocessing module: used to preprocess the data collected by the multidimensional data entry module and transmit the preprocessed data to the indicator calculation module.

[0063] The preprocessing module normalizes the standard deviation of the roasting furnace temperature gradient standard deviation data using a sliding window standard deviation with a window width of 30 seconds, and uses wavelet threshold denoising to eliminate optical interference from the thermal imager; the asphalt smoke emission rate data is smoothed by moving average filtering and then dynamically differentiated with the DCS theoretical value; the number of filler osmotic pressure pulses is tested for Poisson distribution to eliminate random noise and converted into an event rate per unit time; the cooling water temperature difference attenuation coefficient uses the first-order derivative to calculate the temperature difference change slope, and then uses Z-score standardization to eliminate seasonal temperature effects.

[0064] The preprocessing module extracts the 3rd / 5th / 7th harmonic energy proportions after FFT spectrum correction of the current harmonic distortion rate, and uses the Kaiser window function to suppress spectrum leakage; the hydraulic push rod delay data is compensated through PLC clock synchronization, and quartiles are implemented to remove outliers during non-working periods; the vibration frequency band energy proportion is enhanced using an improved algorithm for Mel frequency cepstral coefficients; and the optical fiber strain data is subjected to temperature compensation formula to eliminate thermal expansion pseudo-strain.

[0065] The preprocessing module uses Kriging spatial interpolation to fill the sensor blind spots for the CO concentration variation coefficient and improves data normality through Box-Cox transformation. The electrostatic gradient curve is smoothed by Savitzky-Golay differential method to extract the curvature extreme points. The pressure difference entropy value is calculated using an improved permutation entropy algorithm embedded in the Tau automatic delay selection mechanism. The wind speed anomaly is compressed into a three-dimensional vector through principal component analysis to construct a Mahalanobis distance control chart.

[0066] The preprocessing module performs wavelet packet decomposition on the acoustic emission energy to extract the energy integral of the crack-sensitive frequency band; the near-infrared reflectivity uses the SNV standard normal variable transformation to eliminate surface scattering interference; the dielectric loss angle data is fitted with the Debye model to separate the temperature-frequency cross-influence; and the chromaticity Lab value uses the CIEDE2000 color difference formula to construct an oxidation degree index.

[0067] Index calculation module: used to establish a mathematical model to perform index calculations on the data preprocessed by the preprocessing module, including the process disturbance index calculation model, the equipment health calculation model, the environmental risk value calculation model, and the material degradation index calculation model, and transmit the calculation results to the safety risk assessment module.

[0068] The process disturbance index calculation model is specifically expressed as:

[0069]

[0070] ΔT'_std: normalized standard deviation of temperature gradient, reflecting the uniformity of heat distribution in the furnace;

[0071] Q_dev: The deviation rate between the measured and theoretical values of asphalt flue gas concentration, which represents the abnormal degree of volatilization control;

[0072] λ_pulse: The number of pulses exceeding the osmotic pressure limit per unit time, indicating the stability of the filler;

[0073] α_cool: Normalized cooling efficiency attenuation coefficient, reflecting the degradation state of the heat exchange system.

[0074] The process disturbance index calculation model takes the logarithm of ΔT'_std to eliminate dimensional differences and prevent extreme values from dominating. The square of Q_dev amplifies the influence weight of high concentration deviation. The geometric mean of λ_pulse and α_cool balances the effects of frequency and amplitude. The overall coefficient of 1 / 4 is taken to constrain the output range to the interval [0, 1].

[0075] The device health calculation model is specifically expressed as follows:

[0076] x1 = THD_m: main harmonic distortion rate, k1 = 15%, w1 = 0.4;

[0077] x2 = τ_hyd: hydraulic delay time, k2 = 120ms, w2 = 0.3;

[0078] x3 = E_vib: vibration energy ratio, k3 = 8%, w3 = 0.2;

[0079] x4=ε_real: true material strain, k4=500μm / m, w4=0.1.

[0080] The equipment health calculation model uses a product form to reflect the coupling effect of multiple failure modes of equipment. The benchmark parameter k_i is set to correspond to the industry safety threshold. The index weight w_i is determined based on FMEA analysis to determine the degree of impact. When x_i exceeds k_i, the corresponding factor is greater than 2, triggering nonlinear growth.

[0081] The environmental risk value calculation model is specifically expressed as follows:

[0082]

[0083] CV_co: coefficient of variation of CO concentration, the larger the value, the more uneven the distribution;

[0084] κ_es: extreme value of electrostatic gradient curvature, reflecting the degree of charge accumulation;

[0085] H_p: pressure difference permutation entropy, representing the degree of chaos in the system;

[0086] D_wind: wind speed Mahalanobis distance, the degree to which abnormal airflow deviates from normal operating conditions.

[0087] The environmental risk value calculation model uses the logistic function to map the output to the (0,1) probability space. The coefficients are determined through principal component analysis: the pressure difference entropy has the largest weight (40%), the wind speed anomaly has a negative weight reflecting its inhibitory effect on other parameters, and the exponential term constructs a linear decision boundary, which is converted into risk probability through Sigmoid.

[0088] The material degradation index calculation model is specifically expressed as:

[0089]

[0090] E_ae: crack acoustic emission energy, the larger the value, the more serious the internal defect;

[0091] R_Na: Sodium penetration reflectivity ratio, the smaller the value, the greater the penetration depth;

[0092] tanδ_e: Equivalent dielectric loss angle, reflecting the degree of incomplete coking of the binder;

[0093] ΔE_ox: Oxidation color difference, a large value indicates severe surface oxidation.

[0094] The material degradation index calculation model reduces the absolute dominant effect of energy parameters through cube root processing, and the product form represents the multi-physical field coupling mechanism of material failure. Sodium penetration is used as an independent subtraction term to reflect its compensatory effect on anode conductivity. The empirical coefficient of 0.5 is determined by orthogonal test data regression.

[0095] Security risk assessment module: Based on the calculation results of the indicator calculation module, a multi-level fusion assessment architecture is used to quantify risks.

[0096] The specific process of the security risk assessment module includes:

[0097] S1: Establish a four-dimensional state space:

[0098] is a four-dimensional eigenvector;

[0099] S2: Calculate the degree of deviation between the current state and the historical safety benchmark through Mahalanobis distance: D risk is the risk deviation, is the historical security benchmark vector, T is the transposed vector;

[0100] S3: Risk level classification: Risk levels are divided into L1, L2, L3 and L4, L1 is a blue warning, risk <2σ, trigger, L2 is yellow warning, when 2σ≤D risk <3σ, trigger, L3 is orange warning, when 3σ≤D riskL4 is a red alert and is triggered when D is less than 4σ. L4 is a red alert and is triggered when D ≥ 4σ. σ represents the dynamically updated standard deviation, which is updated every quarter. The sliding window standard deviation is used, and the window = 90 days.

[0101] The present invention first uses a multidimensional data entry module to collect four types of data from the roasting workshop in real time: process disturbance monitoring data, equipment health status data, environmental risk coupling data, and material property derivative data. Special equipment such as infrared thermal imagers, electrochemical sensors, and acoustic emission sensors are used to capture data at specific frequencies and algorithms. Subsequently, a preprocessing module performs denoising, normalization, and spatial interpolation on the raw data. For example, wavelet thresholding is used to eliminate optical interference, Kriging interpolation is used to fill blind spots, and Savitzky-Golay smoothing is used to extract features. The processed data is input into an indicator calculation module, which performs calculations using a process disturbance index model, an equipment health model, an environmental risk value model, and a material degradation index model, respectively. Finally, a safety risk assessment module constructs a four-dimensional state space, calculates the deviation of the current state from a dynamically updated historical safety benchmark based on the Mahalanobis distance, and divides the warning levels into four levels: blue, yellow, orange, and red according to the 2σ, 3σ, and 4σ thresholds to achieve risk quantification output. The standard deviation is updated quarterly through a 90-day sliding window data to ensure the timeliness of the assessment.

[0102] Secondly: The drawings of the embodiments disclosed in the present invention only involve structures related to the embodiments disclosed in the present invention. Other structures may refer to conventional designs. The same embodiment and different embodiments of the present invention may be combined with each other without conflict.

[0103] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A safety risk assessment system for a baking shop for carbon anode production, characterized in that: include: Multi-dimensional data entry module: used to collect process disturbance monitoring data, equipment health status data, environmental risk coupling data, and material property derivative data of the target roasting workshop, and transmit the collected data to the preprocessing module; Preprocessing module: used to preprocess the data collected by the multidimensional data entry module and transmit the preprocessed data to the indicator calculation module; Index calculation module: used to establish a mathematical model to calculate indicators of the data preprocessed by the preprocessing module, including the process disturbance index calculation model, equipment health calculation model, environmental risk value calculation model and material degradation index calculation model, and transmit the calculation results to the safety risk assessment module; Security risk assessment module: Based on the calculation results of the indicator calculation module, a multi-level fusion assessment architecture is used to quantify risks.

2. A safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The process disturbance monitoring data include the standard deviation of the temperature gradient of carbon blocks in the roasting furnace, the asphalt flue gas escape rate, the filler osmotic pressure fluctuation frequency and the cooling water system temperature difference attenuation coefficient; the equipment health status data include the harmonic distortion rate of the ring furnace drive motor current, the hydraulic push rod displacement lag time, the burner nozzle carbon deposition vibration amplitude and the refractory material micro-strain rate; the environmental risk coupling data include the spatial variation coefficient of the CO concentration in the dead corner of the workshop, the electrostatic potential gradient on the ground of the asphalt pump room, the pressure difference fluctuation entropy value of the flue gas purification tower and the instantaneous wind speed vector anomaly in the loading and unloading area; the material property derivative data include the cumulative energy of the acoustic emission of cracks in the anode blank, the infrared reflectivity of the residual sodium penetration depth, the dielectric loss angle of the binder coking process and the Lab value of the oxidation chromaticity of the anode surface out of the furnace.

3. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The pre-processing module normalizes the standard deviation of the roasting furnace temperature gradient standard deviation data using a sliding window with a window width of 30 seconds, and uses wavelet threshold denoising to eliminate optical interference from the thermal imager; the asphalt fume emission rate data is smoothed by a moving average filter and then dynamically differentiated with the DCS theoretical value; The number of filler osmotic pressure pulses was tested using Poisson distribution to eliminate random noise and converted into event rate per unit time. The cooling water temperature difference attenuation coefficient was calculated using the first-order derivative to calculate the temperature difference change slope, and then normalized using Z-score to eliminate seasonal temperature effects. The preprocessing module extracts the 3rd / 5th / 7th harmonic energy proportions after FFT spectrum correction of the current harmonic distortion rate, and uses the Kaiser window function to suppress spectrum leakage; the hydraulic push rod delay data is compensated through PLC clock synchronization, and quartiles are implemented to remove outliers during non-working periods; the vibration frequency band energy proportion is enhanced using an improved algorithm for Mel frequency cepstral coefficients; and the optical fiber strain data is subjected to temperature compensation formula to eliminate thermal expansion pseudo-strain.

4. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The preprocessing module uses Kriging spatial interpolation to fill sensor blind spots for the CO concentration coefficient of variation and improves data normality through Box-Cox transformation. The electrostatic gradient curve is smoothed by Savitzky-Golay differential method and then the curvature extreme points are extracted. The pressure difference entropy value is calculated using an improved permutation entropy algorithm and embedded in the Tau automatic delay selection mechanism. The wind speed anomaly is compressed into a three-dimensional vector through principal component analysis to construct a Mahalanobis distance control chart. The preprocessing module performs wavelet packet decomposition on the acoustic emission energy to extract the energy integral of the crack-sensitive frequency band; the near-infrared reflectivity uses the SNV standard normal variable transformation to eliminate surface scattering interference; the dielectric loss angle data is fitted with the Debye model to separate the temperature-frequency cross-influence; and the chromaticity Lab value uses the CIEDE2000 color difference formula to construct an oxidation degree index.

5. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The process disturbance index calculation model is specifically expressed as: ΔT'_std: normalized standard deviation of temperature gradient, reflecting the uniformity of heat distribution in the furnace; Q_dev: The deviation rate between the measured and theoretical values of asphalt flue gas concentration, which represents the abnormal degree of volatilization control; λ_pulse: The number of pulses exceeding the osmotic pressure limit per unit time, indicating the stability of the filler; α_cool: Normalized cooling efficiency attenuation coefficient, reflecting the degradation state of the heat exchange system.

6. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The device health calculation model is specifically expressed as follows: x1 = THD_m: main harmonic distortion, k1 = 15%, w1 = 0.4; x2 = τ_hyd: hydraulic delay time, k2 = 120ms, w2 = 0.3; x3 = E_vib: vibration energy ratio, k3 = 8%, w3 = 0.2; x4=ε_real: true material strain, k4=500μm / m, w4=0.

1.

7. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The environmental risk value calculation model is specifically expressed as follows: CV_co: coefficient of variation of CO concentration, the larger the value, the more uneven the distribution; κ_es: extreme value of electrostatic gradient curvature, reflecting the degree of charge accumulation; H_p: pressure difference permutation entropy, representing the degree of chaos in the system; D_wind: wind speed Mahalanobis distance, the degree to which abnormal airflow deviates from normal operating conditions.

8. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The material degradation index calculation model is specifically expressed as: E_ae: crack acoustic emission energy, the larger the value, the more serious the internal defect; R_Na: Sodium penetration reflectivity ratio, the smaller the value, the greater the penetration depth; tanδ_e: Equivalent dielectric loss angle, reflecting the degree of incomplete coking of the binder; ΔE_ox: Oxidation color difference, a large value indicates severe surface oxidation.

9. The safety risk assessment system for a baking shop for carbon anode production according to claim 1, characterized in that: The specific process of the security risk assessment module includes: S1: Establish a four-dimensional state space: is a four-dimensional eigenvector; S2: Calculate the degree of deviation between the current state and the historical safety benchmark through Mahalanobis distance: D risk is the risk deviation, is the historical security benchmark vector, T is the transposed vector; S3: Risk level classification: Risk levels are divided into L1, L2, L3 and L4, L1 is a blue warning, risk <2σ, trigger, L2 is yellow warning, when 2σ≤D risk <3σ, trigger, L3 is orange warning, when 3σ≤D risk L4 is a red alert and is triggered when D is less than 4σ. L4 is a red alert and is triggered when D ≥ 4σ. σ represents the dynamically updated standard deviation, which is updated every quarter. The sliding window standard deviation is used, and the window = 90 days.

Images