Processing process fault prediction method for essential oil production

Abstract

The invention discloses a processing process fault prediction method for essential oil production, and relates to the technical field of essential oil processing, and the method comprises the steps: constructing an environment equipment coupling model, and drawing an electrical equipment insulation resistance attenuation curve through a salt mist deposition experiment; environment disturbance variables are introduced into the generative adversarial network, and an anomaly detector capable of recognizing salt spray corrosion characteristics and humidity condensation characteristics is trained; capturing time-space correlation between temperature mutation and quality fluctuation through a self-attention mechanism; calculating a probability density function of the residual service life of the equipment in real time, and automatically correcting a temperature over-limit threshold value based on the residual service life distribution; calculating the contribution degree of each node to the fault through intervention simulation; and constructing a digital twinborn model containing thermodynamic and hydrodynamic characteristics. Through environment dynamic modeling, causal reasoning and digital twinning methods, the key problems of low fault prediction precision, poor traceability efficiency and lack of quantitative support in decision making in essential oil production are solved.

Description

Technical Field

[0001] This invention relates to the field of essential oil processing technology, specifically to a method for predicting processing failures in essential oil production. Background Art

[0002] Aquatic products are highly susceptible to microbial contamination during processing, transportation, and storage, leading to the degradation of their proteins and amino acids, and the emission of a putrid odor, thus affecting the product's taste and nutritional value. Plant essential oils can extend the shelf life of aquatic foods and are of great significance for their preservation. As a natural resource, essential oils are widely available and contain diverse active ingredients, making their application in the field of food preservation of significant theoretical and practical value.

[0003] Essential oil production is mostly located in coastal or humid areas. Salt spray corrosion and humidity condensation accelerate the aging of electrical equipment insulation. However, existing fault prediction methods do not fully consider the impact of dynamic environmental changes on equipment degradation, resulting in large deviations in the prediction of insulation resistance decay. Fault tracing is inefficient, lacks quantitative model support, and makes it difficult to quickly pinpoint the initial cause as either an electrical short circuit due to salt spray deposition or a process interruption caused by pump cavitation. Summary of the Invention

[0004] To solve the above-mentioned technical problems, a method for predicting processing failures in essential oil production is provided. This technical solution addresses the problems mentioned in the background section.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A method for predicting processing failures in essential oil production includes:

[0007] An environmental equipment coupling model was constructed, and the insulation resistance attenuation curve of electrical equipment was plotted through salt spray deposition experiments. Dynamic correction coefficients were developed by combining humidity sensor data.

[0008] Based on adversarial training technology, environmental perturbation variables are introduced into the generative adversarial network to train an anomaly detector that can identify salt spray corrosion characteristics and humidity condensation characteristics.

[0009] A spatiotemporal coordinate alignment framework was designed to map the distillation temperature curve and chromatographic detection results to a unified time axis. A multi-scale feature extractor was developed to capture the spatiotemporal correlation between temperature abrupt changes and quality fluctuations through a self-attention mechanism.

[0010] Establish a degradation status assessment model, calculate the probability density function of the remaining service life of the equipment in real time, and automatically correct the temperature over-limit threshold based on the distribution of the remaining service life.

[0011] Construct a causal graph model that includes the physical relationships between equipment and control logic. Through intervention simulation calculation, identify the initial cause and propagation path of the fault.

[0012] A digital twin model incorporating thermodynamic and fluid dynamic properties is constructed to establish a mapping relationship between processing parameters and equipment failures. The processing parameters include the distillation column temperature gradient field, condenser heat transfer coefficient, pump cavitation margin, and salt spray deposition rate.

[0013] The forecast results are transformed into visual decision nodes, and the output loss and maintenance costs of at least one treatment measure are predicted based on the digital twin model.

[0014] Preferably, the step of establishing a degradation state assessment model, calculating the probability density function of the remaining service life of the equipment in real time, and automatically correcting the temperature over-limit threshold based on the remaining service life distribution specifically includes:

[0015] An Arrhenius degradation model was established based on equipment material fatigue and thermodynamic equations.

[0016] The model is trained using historical fault data, with sensor features as input and degradation status indicators as output.

[0017] The current degradation trajectory is matched with the degradation curve library in the historical fault data for similarity, and the curve with the closest similarity is output. The remaining time of the curve is output as an estimate of the remaining service life of the equipment.

[0018] By using historical data to estimate shape and scale parameters, the probability density function is obtained;

[0019] When a new sensor data point is added, the degradation status index is recalculated, the degradation trajectory is updated, and the probability density function of the device's remaining service life is dynamically adjusted.

[0020] Based on the importance of the equipment and the consequences of failure, a correction factor is preset for the temperature over-limit threshold, and the real-time temperature over-limit threshold is calculated based on the correction factor.

[0021] Preferably, the construction of a causal graph model that includes the physical relationships of equipment and control logic, and the identification of the initial cause and propagation path of the fault by intervening in the simulation calculation of the contribution of each node to the fault, specifically includes:

[0022] Draw a causal network, with the starting node set as the input variable, the intermediate nodes set as variables affected by other nodes, and the result node set as the output fault alarm;

[0023] The arrows point from cause to effect, indicating the relationship type as direct or indirect.

[0024] Design intervention scenarios, conduct single-factor interventions by fixing one variable and observing changes in other variables, and conduct multi-factor joint interventions by fixing at least two relevant variables and observing changes in other variables.

[0025] Simulate fault injection, artificially set component failures, calculate the correlation strength of each node when the fault occurs, and statistically analyze the dependencies between nodes through historical data;

[0026] Change the values ​​of each node in turn, observe the change in the probability of failure, and compare the impact of changes in different nodes on the failure.

[0027] Based on the degree of influence of each node, a weight is assigned to each path. Each weight is recorded as the product of the degree of influence of each node, and the path with the greatest influence from the starting node to the fault node is output.

[0028] Excluding intermediate nodes affected by other nodes, focus on the starting node without a predecessor and count the frequency of occurrence of each starting node in the fault scenario.

[0029] Arrange the starting nodes in descending order of their impact, identify the most important initial factors, and decompose the fault path into a chain structure of initial cause, intermediate event, and fault result.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0031] Insulation resistance decay curves were plotted through salt spray deposition experiments to quantify the relationship between salt spray concentration and insulation aging. Dynamic correction coefficients were developed by combining humidity sensor data to adjust model parameters in real time to adapt to environmental fluctuations. The model's ability to identify salt spray corrosion and humidity condensation was enhanced through adversarial training. A multi-scale feature extractor was developed to capture the spatiotemporal correlation between temperature abrupt changes and quality fluctuations through a self-attention mechanism. Temperature over-limit thresholds were dynamically adjusted to improve the accuracy of temperature over-limit alarms and shorten fault tracing time. Attached Figure Description

[0032] Figure 1 This is a flowchart of the process failure prediction method for essential oil production according to the present invention;

[0033] Figure 2 This is a flowchart of the method for constructing an environmental device coupling model according to the present invention;

[0034] Figure 3 The flowchart of the method for training an anomaly detector capable of recognizing salt spray corrosion characteristics and humidity condensation characteristics according to the present invention is shown below.

[0035] Figure 4 This is a flowchart of the method for developing a multi-scale feature extractor according to the present invention;

[0036] Figure 5The flowchart of the method for establishing a degradation state assessment model of the present invention is shown below;

[0037] Figure 6 The flowchart of the method for constructing a cause-effect graph model including device physical relationships and control logic according to the present invention is shown below;

[0038] Figure 7 This is a flowchart of the method for establishing the mapping relationship between processing parameters and equipment faults according to the present invention. Detailed Implementation

[0039] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0040] Reference Figure 1 As shown, a method for predicting processing failures in essential oil production includes:

[0041] An environmental equipment coupling model was constructed, and the insulation resistance attenuation curve of electrical equipment was plotted through salt spray deposition experiments. Dynamic correction coefficients were developed by combining humidity sensor data.

[0042] Based on adversarial training technology, environmental perturbation variables are introduced into the generative adversarial network to train an anomaly detector that can identify salt spray corrosion characteristics and humidity condensation characteristics.

[0043] A spatiotemporal coordinate alignment framework was designed to map the distillation temperature curve and chromatographic detection results to a unified time axis. A multi-scale feature extractor was developed to capture the spatiotemporal correlation between temperature abrupt changes and quality fluctuations through a self-attention mechanism.

[0044] Establish a degradation status assessment model, calculate the probability density function of the remaining service life of the equipment in real time, and automatically correct the temperature over-limit threshold based on the distribution of the remaining service life.

[0045] Construct a causal graph model that includes the physical relationships between equipment and control logic. Through intervention simulation calculation, identify the initial cause and propagation path of the fault.

[0046] A digital twin model incorporating thermodynamic and fluid dynamic properties is constructed to establish a mapping relationship between processing parameters and equipment failures. The processing parameters include the distillation column temperature gradient field, condenser heat transfer coefficient, pump cavitation margin, and salt spray deposition rate.

[0047] The forecast results are transformed into visual decision nodes, and the output loss and maintenance costs of at least one treatment measure are predicted based on the digital twin model.

[0048] Reference Figure 2As shown, an environmental equipment coupling model is constructed, and the insulation resistance attenuation curve of the electrical equipment is plotted through a salt spray deposition experiment. Dynamic correction coefficients are developed based on humidity sensor data, specifically including:

[0049] A dedicated corrosion test chamber with adjustable temperature, salt spray concentration, and spray cycle is used to simulate at least one environmental condition.

[0050] A resistance measuring device is installed on the sample surface, and a temperature and humidity sensor and a salt spray concentration monitoring device are arranged inside the test chamber.

[0051] Multiple sets of comparative experiments were set up, with salt spray concentrations of 0.5%, 3%, and 5% salt solution, ambient temperatures of 25℃, 40℃, and 60℃, and humidity of 95% to accelerate corrosion.

[0052] Before the experiment, the initial insulation resistance of the sample was measured. Salt spray was started at the set concentration and temperature. The spraying was paused after a fixed time to simulate an intermittent corrosion environment.

[0053] Regularly record sample resistance values, real-time temperature and humidity, and salt spray concentration to continuously cover the long-term decay process;

[0054] Calculate the percentage change based on the initial value and output it as the relative rate of resistance decay. Plot decay curves grouped by temperature and salt spray concentration.

[0055] The half-life and the start time of the steady-state phase in the decay curve are marked as key feature points.

[0056] Collect real-time data from humidity sensors at at least one location on the surface of the device, calculate the average humidity value, calculate the time integral of the hourly average humidity and record it as the cumulative humidity, and calculate the rate of change of humidity.

[0057] By combining the insulation resistance attenuation curve and humidity data from salt spray experiments, a correlation between the dynamic correction coefficient and humidity-related parameters is established.

[0058] Input the current average humidity, cumulative amount, and rate of change. Calculate the dynamic adjustment coefficient based on the correlation, and divide the original resistance prediction value by the adjustment coefficient to obtain the correction result based on the influence of humidity.

[0059] A high-precision resistance measuring device is installed on the surface of the electrical equipment sample to ensure accurate measurement of the insulation resistance value of the sample under different environmental conditions. Temperature and humidity sensors and salt spray concentration monitoring equipment are reasonably arranged inside the test chamber to ensure real-time monitoring and recording of environmental parameters inside the test chamber. Combining the insulation resistance attenuation curve and humidity data obtained from the salt spray experiment, the influence of humidity parameters, including average humidity value, cumulative humidity, and humidity change rate, on insulation resistance attenuation is analyzed.

[0060] Reference Figure 3As shown, based on adversarial training technology, environmental perturbation variables are introduced into the generative adversarial network to train an anomaly detector that can identify salt spray corrosion characteristics and humidity condensation characteristics. Specifically, this includes:

[0061] By simulating the salt spray corrosion and humidity condensation characteristics in real environments through generative adversarial networks, and by using adversarial training between the generator and the discriminator, the discriminator is made capable of recognizing abnormal features.

[0062] The generator produces synthetic data with simulated anomaly features based on random noise and environmental disturbance parameters, including salt spray intensity and humidity level.

[0063] The discriminator learns to distinguish between real and synthetic data and grasps the typical characteristics of salt spray corrosion and humidity condensation, and is used as an anomaly detector.

[0064] Collect sensor data from industrial equipment and metal structures in salt spray and high humidity environments, and label corrosion spots and water condensation marks in the data;

[0065] Random noise is generated and combined with parameters of salt spray intensity and humidity level. The resulting data is then input into the generator to generate synthetic data. The real data and synthetic data are then input into the discriminator to calculate the discrimination result.

[0066] Adjust the discriminator parameters based on the discrimination results;

[0067] Train the generator, fix the discriminator parameters, and adjust the generator parameters until the discriminator can no longer distinguish between real data and synthetic data.

[0068] The generator receives two sets of inputs: random noise sampled from a Gaussian or uniform distribution as the basis for data diversity; environmental disturbance parameters including salt spray intensity and humidity level; and a network structure using fully connected or convolutional layers to map noise and environmental parameters to the latent space. Synthetic data with dimensions consistent with real sensor data is generated through transposed convolution or upsampling layers. The output is synthetic data with simulated salt spray corrosion such as surface pitting, oxide layers, or humidity condensation such as water mist condensation marks and humidity gradient features.

[0069] Reference Figure 4 As shown, a spatiotemporal coordinate alignment framework is designed to map the distillation temperature curve and chromatographic detection results to a unified time axis. A multi-scale feature extractor is developed to capture the spatiotemporal correlation between temperature abrupt changes and quality fluctuations through a self-attention mechanism. Specifically, this includes:

[0070] Using the temperature abrupt change point and the time of appearance of the chromatographic peak during the distillation process as anchor points, the time axes of the two types of data are segmented and aligned;

[0071] The continuous time stream was divided into fixed-length time windows, and the mean temperature and mass distribution of chromatographic components were statistically analyzed within each window.

[0072] Design a multi-scale feature extractor, input aligned time window data, use a one-dimensional convolutional network to extract temperature mutation fragments and chromatographic peak fluctuations, and record them as fine-grained feature vectors;

[0073] Based on the fixed-length time window, an original time window is set, two adjacent original time windows are merged to form a second-level window, two adjacent second-level windows are merged to form a third-level window, and so on, until the entire time series length is covered.

[0074] For each level of window feature, the maximum value of the data within the window is taken as the representative value, and the aggregation results of all levels are concatenated to form a coarse-grained feature vector.

[0075] Fine-grained and coarse-grained feature vectors are concatenated to form a multi-scale feature representation;

[0076] A linear transformation is performed on the feature sequence to generate a query, key, and value matrix. The dot product of the query and the key is calculated, and the attention score reflecting the association strength between different time steps is obtained through normalization.

[0077] The attention scores are used to weight and sum the value matrix to generate context-aware feature representations.

[0078] In the attention mechanism, the forced temperature change rate feature is used as the query vector to match abnormal fluctuations in the quality of chromatographic components;

[0079] By introducing time delay embedding, the hysteretic effect of modeling temperature changes on quality fluctuations is revealed.

[0080] Using the abrupt change points in the distillation temperature curve and the start and peak times of the chromatographic peaks in the chromatographic detection results as alignment anchors, the time axis of the temperature curve and chromatographic data is divided into corresponding segments according to the anchors. This ensures that the temperature abrupt change and the time range of the chromatographic peak strictly correspond within each segment. Linear interpolation is then performed on the segmented temperature data and chromatographic data to unify the time resolution and eliminate time sampling differences.

[0081] Reference Figure 5 As shown, a degradation state assessment model is established to calculate the probability density function of the remaining service life of the equipment in real time, and to automatically correct the temperature over-limit threshold based on the remaining service life distribution. Specifically, this includes:

[0082] An Arrhenius degradation model was established based on equipment material fatigue and thermodynamic equations.

[0083] The model is trained using historical fault data, with sensor features as input and degradation status indicators as output.

[0084] The current degradation trajectory is matched with the degradation curve library in the historical fault data for similarity, and the curve with the closest similarity is output. The remaining time of the curve is output as an estimate of the remaining service life of the equipment.

[0085] By using historical data to estimate shape and scale parameters, the probability density function is obtained;

[0086] When a new sensor data point is added, the degradation status index is recalculated, the degradation trajectory is updated, and the probability density function of the device's remaining service life is dynamically adjusted.

[0087] Based on the importance of the equipment and the consequences of failure, a correction factor is preset for the temperature over-limit threshold, and the real-time temperature over-limit threshold is calculated based on the correction factor.

[0088] The Arrhenius degenerate model is as follows:

[0089] ,

[0090] In the formula, The degradation rate constant represents the degradation rate of the equipment material at the rated temperature. This is a frequency factor, related to the material properties of the equipment. The apparent activation energy represents the energy required for the degradation reaction. For molar gas constant, Absolute temperature;

[0091] The probability density function is:

[0092] ,

[0093] In the formula, The remaining useful life time is the variable. The shape parameter controls the shape of the distribution curve. This is a scale parameter representing the characteristic lifetime, i.e., 63.2% of the equipment fails within this time.

[0094] The real-time temperature exceedance threshold is:

[0095] ,

[0096] In the formula, For real-time temperature exceeding the threshold, As the initial threshold, This is a correction factor.

[0097] Reference Figure 6 As shown, a causal graph model incorporating equipment physical relationships and control logic is constructed. Through intervention simulation calculations, the contribution of each node to the fault is identified, and the initial causes and propagation paths of the fault are determined. Specifically, this includes:

[0098] Draw a causal network, with the starting node set as the input variable, the intermediate nodes set as variables affected by other nodes, and the result node set as the output fault alarm;

[0099] The arrows point from cause to effect, indicating the relationship type as direct or indirect.

[0100] Design intervention scenarios, conduct single-factor interventions by fixing one variable and observing changes in other variables, and conduct multi-factor joint interventions by fixing at least two relevant variables and observing changes in other variables.

[0101] Simulate fault injection, artificially set component failures, calculate the correlation strength of each node when the fault occurs, and statistically analyze the dependencies between nodes through historical data;

[0102] Change the values ​​of each node in turn, observe the change in the probability of failure, and compare the impact of changes in different nodes on the failure.

[0103] Based on the degree of influence of each node, a weight is assigned to each path. Each weight is recorded as the product of the degree of influence of each node, and the path with the greatest influence from the starting node to the fault node is output.

[0104] Excluding intermediate nodes affected by other nodes, focus on the starting node without a predecessor and count the frequency of occurrence of each starting node in the fault scenario.

[0105] Arrange the starting nodes in descending order of their impact, identify the most important initial factors, and decompose the fault path into a chain structure of initial cause, intermediate event, and fault result.

[0106] The starting node is the input variable, such as equipment operating parameters like temperature setpoint and flow rate, environmental variables like raw material humidity, and sensor readings like pressure and liquid level. The intermediate node is the process variable, such as variables affected by other nodes like actual temperature, heat exchanger efficiency, and valve opening. The result node is the fault alarm, which outputs fault flags such as overheat alarm, pressure over-limit, and abnormal chromatographic peak. The arrow direction points from cause to result. The relationship type is indicated: direct effect is physical connection or strong logical association, and indirect effect is the influence transmitted through intermediate variables.

[0107] Reference Figure 7 As shown, constructing a digital twin model incorporating thermodynamic and fluid dynamic properties and establishing a mapping relationship between processing parameters and equipment failures specifically includes:

[0108] The distillation column is divided into at least one thermodynamic unit, and the temperature, pressure and component distribution of each layer are independently modeled. The boundary conditions of each layer are dynamically adjusted by processing real-time sensor data to simulate transient thermal response. The real-time sensor data includes feed temperature and heating power.

[0109] By combining the gas-liquid two-phase flow model, the influence of the velocity distribution in the tower on the temperature gradient is calculated. The dynamic equation of the overall heat transfer coefficient is established by integrating the convective heat transfer on the refrigerant side, the heat conduction on the metal wall and the heat transfer on the cooling medium side.

[0110] The mineralization of cooling water is detected by a conductivity sensor, and the fouling thickness is predicted by combining flow velocity data to dynamically correct the heat transfer coefficient.

[0111] Simulate the pressure distribution inside the pump, identify low-pressure areas and calculate net positive suction head (NPSH), and correct the allowable NPSH in real time based on liquid temperature, altitude and suction line resistance.

[0112] By combining salt spray concentration, humidity and substrate surface energy, the deposition rate of salt spray deposition is calculated, and the film thickness is monitored in real time by a laser interferometer to adjust the deposition rate threshold.

[0113] Based on the mathematical relationship between temperature gradient field and separation efficiency, when the temperature of a certain layer deviates from the set value, a warning of decreased separation efficiency is triggered, and a distillation column fault signal is output.

[0114] Determine if the heat transfer coefficient has decreased by 15%. If so, it is related to insufficient cooling water flow or dirt accumulation, triggering a cleaning command and outputting a condenser fault signal. If not, no output is made.

[0115] Determine if the real-time NPSH is lower than the allowable NPSH. If so, map it to cavitation risk, automatically adjust the operating frequency, and output a pump fault signal. If not, no output is made.

[0116] Determine if the deposition rate exceeds the deposition rate threshold. If so, record the decrease in the film's salt spray resistance, trigger a maintenance reminder, and output a salt spray fault signal. If not, no output is made.

[0117] The distillation column is divided into multiple thermodynamic units along its height, with each unit being 1 meter high. Mathematical relationships for temperature, pressure, and component distribution are established for each unit. The number of units can be adjusted according to actual processing requirements. Feed temperature and heating power data are acquired in real time by sensors and used as input conditions for each unit's model. Based on the principles of material and energy balance, the changes in temperature, pressure, and components are calculated layer by layer, and the mass and heat transfer parameters between units are updated. Numerical calculation methods such as the finite difference method or the finite volume method are used to simulate the temperature change of each unit over time, reflecting the transient thermal behavior during heating or cooling.

[0118] Furthermore, this solution also proposes a computer-readable storage medium storing a computer-readable program, which, when invoked, executes the aforementioned method for predicting processing failures in essential oil production.

[0119] It is understandable that the storage medium can be a magnetic medium, such as a floppy disk, hard disk, or magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid-state drive (SSD).

[0120] In summary, the advantages of this invention are as follows: by plotting insulation resistance decay curves through salt spray deposition experiments, the relationship between salt spray concentration and insulation aging is quantified; dynamic correction coefficients are developed in conjunction with humidity sensor data, and model parameters are adjusted in real time to adapt to environmental fluctuations; adversarial training enhances the model's ability to identify salt spray corrosion and humidity condensation; a multi-scale feature extractor is developed; a self-attention mechanism is used to capture the spatiotemporal correlation between temperature mutations and quality fluctuations; temperature over-limit thresholds are dynamically adjusted, improving the accuracy of temperature over-limit alarms and shortening fault tracing time.

[0121] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for predicting processing failures in essential oil production, characterized in that, include: An environmental equipment coupling model was constructed, and the insulation resistance attenuation curve of electrical equipment was plotted through salt spray deposition experiments. Dynamic correction coefficients were developed by combining humidity sensor data. Based on adversarial training technology, environmental perturbation variables are introduced into the generative adversarial network to train an anomaly detector that can identify salt spray corrosion characteristics and humidity condensation characteristics. A spatiotemporal coordinate alignment framework was designed to map the distillation temperature curve and chromatographic detection results to a unified time axis. A multi-scale feature extractor was developed to capture the spatiotemporal correlation between temperature abrupt changes and quality fluctuations through a self-attention mechanism. Establish a degradation status assessment model, calculate the probability density function of the remaining service life of the equipment in real time, and automatically correct the temperature over-limit threshold based on the distribution of the remaining service life. Construct a causal graph model that includes the physical relationships between equipment and control logic. Through intervention simulation calculation, identify the initial cause and propagation path of the fault. A digital twin model incorporating thermodynamic and fluid dynamic properties is constructed to establish a mapping relationship between processing parameters and equipment failures. The processing parameters include the distillation column temperature gradient field, condenser heat transfer coefficient, pump NPSH, and salt spray deposition rate. The forecast results are transformed into visual decision nodes, and the output loss and maintenance costs of at least one treatment measure are predicted based on the digital twin model.

2. The method for predicting processing failures in essential oil production according to claim 1, characterized in that, The construction of the environmental equipment coupling model, the plotting of the insulation resistance attenuation curve of electrical equipment through salt spray deposition experiments, and the development of dynamic correction coefficients based on humidity sensor data specifically include: A dedicated corrosion test chamber with adjustable temperature, salt spray concentration, and spray cycle is used to simulate at least one environmental condition. A resistance measuring device is installed on the sample surface, and a temperature and humidity sensor and a salt spray concentration monitoring device are arranged inside the test chamber. Multiple sets of comparative experiments were set up, with salt spray concentrations of 0.5%, 3%, and 5% salt solution, ambient temperatures of 25℃, 40℃, and 60℃, and humidity of 95% to accelerate corrosion. Before the experiment, the initial insulation resistance of the sample was measured. Salt spray was started at the set concentration and temperature. The spraying was paused after a fixed time to simulate an intermittent corrosion environment. Regularly record sample resistance values, real-time temperature and humidity, and salt spray concentration to continuously cover the long-term decay process; Calculate the percentage change based on the initial value and output it as the relative rate of resistance decay. Plot decay curves grouped by temperature and salt spray concentration. The half-life and the start time of the steady-state phase in the decay curve are marked as key feature points. Collect real-time data from humidity sensors at at least one location on the surface of the device, calculate the average humidity value, calculate the time integral of the hourly average humidity and record it as the cumulative humidity, and calculate the rate of change of humidity. By combining the insulation resistance attenuation curve and humidity data from salt spray experiments, a correlation between the dynamic correction coefficient and humidity-related parameters is established. Input the current average humidity, cumulative amount, and rate of change. Calculate the dynamic adjustment coefficient based on the correlation, and divide the original resistance prediction value by the adjustment coefficient to obtain the correction result based on the influence of humidity.

3. The method for predicting processing failures in essential oil production according to claim 2, characterized in that, The aforementioned adversarial training technique, which introduces environmental perturbation variables into the generative adversarial network to train an anomaly detector capable of recognizing salt spray corrosion characteristics and humidity condensation characteristics, specifically includes: By simulating the salt spray corrosion and humidity condensation characteristics in real environments through generative adversarial networks, and by using adversarial training between the generator and the discriminator, the discriminator is made capable of recognizing abnormal features. The generator produces synthetic data with simulated anomaly features based on random noise and environmental disturbance parameters, including salt spray intensity and humidity level. The discriminator learns to distinguish between real and synthetic data and grasps the typical characteristics of salt spray corrosion and humidity condensation, and is used as an anomaly detector. Collect sensor data from industrial equipment and metal structures in salt spray and high humidity environments, and label corrosion spots and water condensation marks in the data; Random noise is generated and combined with parameters of salt spray intensity and humidity level. The resulting data is then input into the generator to generate synthetic data. The real data and synthetic data are then input into the discriminator to calculate the discrimination result. Adjust the discriminator parameters based on the discrimination results; Train the generator, fix the discriminator parameters, and adjust the generator parameters until the discriminator can no longer distinguish between real data and synthetic data.

4. The method for predicting processing failures in essential oil production according to claim 3, characterized in that, The design of the spatiotemporal coordinate alignment framework maps the distillation temperature curve and chromatographic detection results to a unified time axis. A multi-scale feature extractor is developed, and a self-attention mechanism is used to capture the spatiotemporal correlation between temperature abrupt changes and quality fluctuations. Specifically, this includes: Using the temperature abrupt change point and the time of appearance of the chromatographic peak during the distillation process as anchor points, the time axes of the two types of data are segmented and aligned; The continuous time stream was divided into fixed-length time windows, and the mean temperature and mass distribution of chromatographic components were statistically analyzed within each window. Design a multi-scale feature extractor, input aligned time window data, use a one-dimensional convolutional network to extract temperature mutation fragments and chromatographic peak fluctuations, and record them as fine-grained feature vectors; Based on the fixed-length time window, an original time window is set, two adjacent original time windows are merged to form a second-level window, two adjacent second-level windows are merged to form a third-level window, and so on, until the entire time series length is covered. For each level of window feature, the maximum value of the data within the window is taken as the representative value, and the aggregation results of all levels are concatenated to form a coarse-grained feature vector. Fine-grained and coarse-grained feature vectors are concatenated to form a multi-scale feature representation; A linear transformation is performed on the feature sequence to generate a query, key, and value matrix. The dot product of the query and the key is calculated, and the attention score reflecting the association strength between different time steps is obtained through normalization. The attention scores are used to weight and sum the value matrix to generate context-aware feature representations. In the attention mechanism, the forced temperature change rate feature is used as the query vector to match abnormal fluctuations in the quality of chromatographic components; By introducing time delay embedding, the hysteretic effect of modeling temperature changes on quality fluctuations is revealed.

5. A method for predicting processing faults in essential oil production according to claim 4, characterized in that, The establishment of a degradation state assessment model, the real-time calculation of the probability density function of the remaining service life of the equipment, and the automatic correction of the temperature over-limit threshold based on the remaining service life distribution specifically include: An Arrhenius degradation model was established based on equipment material fatigue and thermodynamic equations. The model is trained using historical fault data, with sensor features as input and degradation status indicators as output. The current degradation trajectory is matched with the degradation curve library in the historical fault data for similarity, and the curve with the closest similarity is output. The remaining time of the curve is output as an estimate of the remaining service life of the equipment. By using historical data to estimate shape and scale parameters, the probability density function is obtained; When a new sensor data point is added, the degradation status index is recalculated, the degradation trajectory is updated, and the probability density function of the device's remaining service life is dynamically adjusted. Based on the importance of the equipment and the consequences of failure, a correction factor is preset for the temperature over-limit threshold, and the real-time temperature over-limit threshold is calculated based on the correction factor.

6. The method for predicting processing failures in essential oil production according to claim 5, characterized in that, The construction of a causal graph model that includes the physical relationships of equipment and control logic, and the identification of the initial cause and propagation path of the fault by intervening in the simulation calculation of the contribution of each node to the fault, specifically includes: Draw a causal network, with the starting node set as the input variable, the intermediate nodes set as variables affected by other nodes, and the result node set as the output fault alarm; The arrows point from cause to effect, indicating the relationship type as direct or indirect. Design intervention scenarios, conduct single-factor interventions by fixing one variable and observing changes in other variables, and conduct multi-factor joint interventions by fixing at least two relevant variables and observing changes in other variables. Simulate fault injection, artificially set component failures, calculate the correlation strength of each node when the fault occurs, and statistically analyze the dependencies between nodes through historical data; Change the values ​​of each node in turn, observe the change in the probability of failure, and compare the impact of changes in different nodes on the failure. Based on the degree of influence of each node, a weight is assigned to each path. Each weight is recorded as the product of the degree of influence of each node, and the path with the greatest influence from the starting node to the fault node is output. Excluding intermediate nodes affected by other nodes, focus on the starting node without a predecessor and count the frequency of occurrence of each starting node in the fault scenario. Arrange the starting nodes in descending order of their impact, identify the most important initial factors, and decompose the fault path into a chain structure of initial cause, intermediate event, and fault result.

7. A method for predicting processing failures in essential oil production according to claim 6, characterized in that, The construction of a digital twin model incorporating thermodynamic and hydrodynamic properties, and the establishment of a mapping relationship between processing parameters and equipment failures, specifically includes: The distillation column is divided into at least one thermodynamic unit, and the temperature, pressure and component distribution of each layer are independently modeled. The boundary conditions of each layer are dynamically adjusted by processing real-time sensor data to simulate transient thermal response. The real-time sensor data includes feed temperature and heating power. By combining the gas-liquid two-phase flow model, the influence of the velocity distribution in the tower on the temperature gradient is calculated. The dynamic equation of the overall heat transfer coefficient is established by integrating the convective heat transfer on the refrigerant side, the heat conduction on the metal wall and the heat transfer on the cooling medium side. The mineralization of cooling water is detected by a conductivity sensor, and the fouling thickness is predicted by combining flow velocity data to dynamically correct the heat transfer coefficient. Simulate the pressure distribution inside the pump, identify low-pressure areas and calculate net positive suction head (NPSH), and correct the allowable NPSH in real time based on liquid temperature, altitude and suction line resistance. By combining salt spray concentration, humidity and substrate surface energy, the deposition rate of salt spray deposition is calculated, and the film thickness is monitored in real time by a laser interferometer to adjust the deposition rate threshold. Based on the mathematical relationship between temperature gradient field and separation efficiency, when the temperature of a certain layer deviates from the set value, a warning of decreased separation efficiency is triggered, and a distillation column fault signal is output. Determine if the heat transfer coefficient has decreased by 15%. If so, it is related to insufficient cooling water flow or dirt accumulation, triggering a cleaning command and outputting a condenser fault signal. If not, no output is made. Determine if the real-time NPSH is lower than the allowable NPSH. If so, map it to cavitation risk, automatically adjust the operating frequency, and output a pump fault signal. If not, no output is made. Determine if the deposition rate exceeds the deposition rate threshold. If so, record the decrease in the film's salt spray resistance, trigger a maintenance reminder, and output a salt spray fault signal. If not, no output is made.

Images