Metallurgical furnace fault diagnosis system and method based on FL-CNN-Transform-cGAN

Abstract

The invention discloses a metallurgical furnace fault diagnosis system and method based on FL-CNN-Transform-cGAN. The system comprises an acquisition module, a processing module, a construction module and a diagnosis module. The acquisition module is used for acquiring multi-dimensional data in the operation process of the metallurgical furnace in real time; the processing module is used for preprocessing the multi-dimensional data to obtain processed data; the construction module is used for constructing a diagnosis model based on the processed data; and the diagnosis module is used for completing diagnosis of metallurgical furnace faults by using the diagnosis model. According to the invention, through multi-dimensional data fusion and in combination with an advanced machine learning technology, various fault types and severity can be identified more accurately. In addition, a traditional method generally needs a large amount of historical fault data for model training, and for new equipment or rare faults, the diagnostic performance is limited. According to the method, the simulation fault data is generated by using the cGAN, so that the problem is solved, and the adaptability and robustness of the model are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a metallurgical furnace fault diagnosis system and method based on FL-CNN-Transformer-cGAN. Background Art

[0002] In the modern metallurgical industry, furnaces are core equipment, and their operating status is directly related to production efficiency, product quality, and safety. With the advancement of industrial automation, the requirements for fault diagnosis technology for complex industrial equipment are increasing. Traditional fault diagnosis methods are increasingly unable to meet the needs of refined and intelligent monitoring of furnace equipment under harsh operating conditions such as high temperatures and high loads.

[0003] In recent years, the application of artificial intelligence technology in the industrial sector has made significant progress, with various machine learning models widely used in equipment fault diagnosis. Among them, CNN, with its powerful feature extraction capabilities, excels in processing image and time series data. The Transformer architecture, through its self-attention mechanism, effectively captures long-term dependencies in data, achieving breakthroughs in natural language processing and time series analysis. cGAN, through adversarial training of generators and discriminators, can generate realistic simulated data, providing a new approach to addressing the shortage of actual fault data.

[0004] Federated learning, an emerging distributed machine learning framework, enables joint optimization of model parameters across multiple clients while protecting data privacy. This allows clients to collectively improve model performance without sharing raw data. This feature is crucial for industries with high data privacy requirements, such as the metallurgical industry. Summary of the Invention

[0005] In order to solve the technical problems in the above background, the purpose of the present invention is to provide a metallurgical furnace fault diagnosis system and method based on multi-model fusion. By installing multiple sensors on the outside of the furnace and using federated learning combined with advanced machine learning technologies such as CNN, Transformer and cGAN, real-time monitoring of the furnace operation status, fault prediction and classification can be achieved, thereby improving the reliability and safety of the furnace operation.

[0006] To achieve the above objectives, the present invention provides a metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN, comprising: an acquisition module, a processing module, a construction module and a diagnosis module;

[0007] The acquisition module is used to collect multi-dimensional data during the operation of the metallurgical furnace in real time;

[0008] The processing module is used to pre-process the multi-dimensional data to obtain processed data;

[0009] The construction module is used to construct a diagnostic model based on the processed data;

[0010] The diagnostic module is used to complete the diagnosis of metallurgical furnace faults using the diagnostic model.

[0011] Preferably, the acquisition module includes: a vibration sensor, a temperature sensor, and a pressure sensor;

[0012] The multi-dimensional data collected includes vibration signals of the metallurgical furnace during operation, temperature data at different locations inside the furnace, and internal pressure data;

[0013] The acquisition module is connected to the processing module via a wired or wireless manner.

[0014] Preferably, the vibration signal includes: vibration amplitude, vibration frequency and vibration phase;

[0015] The temperature data includes: the temperature value and the uniformity of temperature distribution;

[0016] The pressure data includes: the size of the pressure value, the frequency of pressure fluctuations and the amplitude of pressure fluctuations.

[0017] Preferably, the processing module uses a low-pass filter to remove high-frequency noise in the vibration signal;

[0018] The processing module normalizes the temperature data and the pressure data, and reduces the dimension of the temperature data and the pressure data by using a principal component analysis technique.

[0019] Preferably, the diagnostic model includes: a CNN model, a Transformer model, a cGAN model and a fully connected layer;

[0020] The CNN model is used to extract data features of the processed data;

[0021] The Transformer model is used to capture the long-term dependencies of the processed data;

[0022] The cGAN model is used to generate a simulated fault data augmentation dataset;

[0023] The fully connected layer is used for fault classification.

[0024] Preferably, the diagnostic model is trained using a federated learning framework, and different clients are used to train the CNN model, the Transformer model, and the cGAN model;

[0025] Finally, the server aggregates the client model parameters to generate a global model, and the global model is used as the diagnosis model.

[0026] The present invention also provides a metallurgical furnace fault diagnosis method based on FL-CNN-Transformer-cGAN, which is applied to the above system and includes the following steps:

[0027] Real-time collection of multi-dimensional data during the operation of metallurgical furnaces;

[0028] Preprocessing the multi-dimensional data to obtain processed data;

[0029] constructing a diagnostic model based on the processed data;

[0030] The diagnosis model is used to diagnose metallurgical furnace faults.

[0031] Preferably, the diagnostic model includes: a CNN model, a Transformer model, a cGAN model and a fully connected layer;

[0032] The CNN model is used to extract data features of the processed data;

[0033] The Transformer model is used to capture the long-term dependencies of the processed data;

[0034] The cGAN model is used to generate a simulated fault data augmentation dataset;

[0035] The fully connected layer is used for fault classification.

[0036] Preferably, the diagnostic model is trained using a federated learning framework, and different clients are used to train the CNN model, the Transformer model, and the cGAN model;

[0037] Finally, the server aggregates the client model parameters to generate a global model, and the global model is used as the diagnosis model.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] This invention uses multi-dimensional data fusion combined with advanced machine learning techniques to more accurately identify various fault types and severity levels. Traditional methods typically require extensive historical fault data for model training, which limits diagnostic performance for new equipment or rare faults. This invention effectively addresses this issue by using cGAN to generate simulated fault data, improving the model's adaptability and robustness.

[0040] The implementation of this invention will significantly improve the operational management of metallurgical furnaces. Through real-time monitoring and early warning of faults, companies can plan maintenance plans in advance, reduce downtime caused by sudden failures, and lower repair costs. Furthermore, accurate fault diagnosis and classification will help extend the service life of furnace equipment, improve production efficiency and product quality, and enhance companies' market competitiveness.

[0041] The multi-model fusion architecture proposed in this paper is not only applicable to metallurgical furnace fault diagnosis, but can also be expanded to fault monitoring and intelligent maintenance for other complex industrial equipment. By analyzing the operational data characteristics of different equipment and adjusting model parameters, this architecture can provide efficient and accurate fault diagnosis solutions for various industrial equipment, promoting the development of industrial intelligence.

[0042] The promotion and application of this invention will bring significant economic and social benefits to the metallurgical industry. By reducing downtime caused by furnace failures, lowering maintenance costs, and improving production efficiency and product quality, enterprises can achieve direct economic benefits. Furthermore, the system's intelligent fault diagnosis and early warning capabilities will effectively improve production safety, reduce the occurrence of safety accidents, and protect the lives of employees and the safety of corporate assets, thus having important social significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0044] Figure 1 Schematic diagram of the system architecture of an embodiment of the present invention;

[0045] Figure 2 This is a method flow framework diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0046] 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.

[0047] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Example 1

[0049] This embodiment provides a metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN, including: an acquisition module, a processing module, a construction module, and a diagnosis module; the acquisition module is used to collect multi-dimensional data during the operation of the metallurgical furnace in real time; the processing module is used to preprocess the multi-dimensional data to obtain processed data; the construction module is used to build a diagnostic model based on the processed data; and the diagnosis module is used to complete the diagnosis of metallurgical furnace faults using the diagnostic model.

[0050] The following will describe in detail how the present invention solves technical problems in practical work in conjunction with this embodiment.

[0051] First, the acquisition module is used to collect multi-dimensional data (raw data) during the operation of the metallurgical furnace in real time. In this embodiment, the acquisition module includes vibration sensors, temperature sensors, and pressure sensors; these sensors are installed at key positions outside the metallurgical furnace to collect multi-dimensional data during the operation of the furnace in real time. The layout of the various sensors in the acquisition module needs to be optimized according to the structure and operation characteristics of the furnace to ensure the comprehensiveness and representativeness of the collected data. For example, vibration sensors should be installed in key parts of the furnace where abnormal vibration is prone to occur, such as furnace body support points, key transmission components, etc.; temperature sensors need to cover the main heating areas and heat exchange parts in the furnace; pressure sensors are installed in areas sensitive to pressure changes in the furnace, such as air inlets and outlets. Through reasonable layout, all-round monitoring of the operation status of the furnace can be achieved.

[0052] Specifically, vibration sensors are used to collect vibration signals from the furnace during operation, including information such as vibration amplitude, vibration frequency, and vibration phase. These data can reflect the mechanical vibration state of the furnace, such as abnormal vibration caused by uneven material distribution inside the furnace, eccentric loading, loose or damaged supporting structures, etc. Temperature sensors are used to monitor temperature data at different locations inside the furnace, including temperature values ​​and the uniformity of temperature distribution. Temperature data is important for determining whether the furnace's heating system is operating normally, whether the insulation material is damaged, and whether the heat transfer inside the furnace is uniform. Pressure sensors are used to detect pressure data inside the furnace, such as the pressure value, the frequency and amplitude of pressure fluctuations, etc. Pressure data can help identify problems such as the furnace's sealing performance and whether the air intake and exhaust systems are operating normally.

[0053] The above raw data has the following characteristics:

[0054] (1) Multidimensionality: Since various types of sensor data such as vibration, temperature and pressure are collected, these data reflect the operating status of the furnace from different angles, enabling the system to have a more comprehensive understanding of the operation of the furnace.

[0055] (2) Real-time: The sensor collects data in real time and can promptly reflect changes in the operating status of the furnace, providing a basis for early warning and real-time diagnosis of faults.

[0056] (3) Complexity: The data generated during the operation of the kiln often contain a lot of noise and interference, such as high-frequency noise in the vibration signal, short-term fluctuations in temperature and pressure data, etc. These need to be removed and corrected through subsequent data preprocessing steps.

[0057] (4) High dimensionality: The original data has a high dimensionality, and directly using it for model training may increase the computational complexity. Therefore, it is necessary to use dimensionality reduction technology to extract key features and reduce the data dimension.

[0058] Afterwards, the processing module preprocesses the multidimensional data to obtain processed data. A low-pass filter is used to remove high-frequency noise from the vibration signal, the temperature and pressure data are normalized, and principal component analysis (PCA) is used to reduce the data's dimensionality. In addition to using a low-pass filter to remove high-frequency noise from the vibration signal, appropriate filters can be selected based on the actual data characteristics. For example, a bandpass filter can retain signals in specific frequency bands to better extract fault characteristics. Temperature and pressure data can be smoothed using methods such as moving averages and median filters to effectively reduce data fluctuations and the impact of outliers.

[0059] In the above normalization process, let the original data be x and the normalized data be x′, then the normalization formula is:

[0060]

[0061] Among them, min(x) and max(x) are the minimum and maximum values ​​of the data x, respectively. This formula is used to map the data to the interval [0,1] to facilitate subsequent model processing.

[0062] The above-mentioned PCA is an important technology for dimensionality reduction. Its core is to find the main characteristic components of the data through eigenvalue decomposition. Let the data matrix be X, and its covariance matrix is:

[0063]

[0064] Here, n0 represents the number of samples.

[0065] Perform eigenvalue decomposition on ∑ and obtain eigenvalues ​​λ1≥λ2≥…≥λ m and the corresponding eigenvectors v1, v2, ..., v m , select the first k eigenvectors to form the projection matrix V, and project the data into a low-dimensional space:

[0066] X reduced =XV

[0067] Among them, X reduced For data after dimensionality reduction, this method can effectively extract the main features of the data, reduce the data dimension and computational complexity.

[0068] A diagnostic model is constructed using the building blocks, and the processed data is fed into the diagnostic model for training. In this embodiment, the diagnostic model uses federated learning combined with models such as CNN, Transformer, and cGAN to train the preprocessed data. Real-time furnace operation data is fed into the trained model to perform fault diagnosis and classification, outputting the fault type, severity, and corresponding treatment recommendations. The model can also provide feedback and optimization based on actual fault handling results and new fault data, continuously improving diagnostic accuracy and adaptability.

[0069] The diagnostic model is trained using a federated learning framework, using different clients to train CNN models, Transformer models, and cGAN models. Finally, the server aggregates the client model parameters to generate a global model, which is used as the diagnostic model. The number of clients and data distribution in the federated learning framework have a significant impact on the performance of the global model. The data from each client should have a certain degree of diversity to cover the different operating conditions and fault types of the furnace. At the same time, the amount of data from each client must be balanced to avoid bias in the global model training caused by excessive or insufficient data from individual clients. When aggregating client model parameters, in addition to using a simple weighted average, the server can also perform weighting based on indicators such as client data quality and model performance to further improve the accuracy and generalization ability of the global model.

[0070] In the federated learning framework, the aggregation of model parameters of each client is a key step. Assume that there are K 总 clients, the model parameter of the kth client is θ k , the amount of data is n k The total amount of data is Then the server aggregates the global model parameter θ global The formula is:

[0071]

[0072] Through this weighted average method, the model parameters of each client are integrated to generate a global model with better generalization ability.

[0073] In the diagnostic model, the CNN model is used to extract data features, and its convolution operation is the core. Its formula is: Let the input feature map be The convolution kernel is Where H and W are the height and width of the feature map, C is the number of channels, and K H , K Wis the height and width of the convolution kernel, and D is the number of convolution kernels. After the convolution operation, the output feature map is obtained in:

[0074]

[0075] Among them, b d is a bias term, which can extract local features of data through convolution operation, providing a basis for subsequent fault diagnosis; m represents the row offset of the convolution kernel on the input feature map; n represents the column offset of the convolution kernel on the input feature map; c1 represents the channel index of the input feature map; Y i,j,d Represents the value of the dth channel at the output feature map at position (i, j); Represents the value of the c1th channel at position (i+m,j+n) of the input feature map; Represents the weight of the convolution kernel at position (m,n) from the c1th input channel to the dth output channel.

[0076] To further improve the efficiency and accuracy of CNN models in feature extraction, this paper introduces Depthwise Separable Convolution. Traditional convolution operations are computationally intensive when processing high-dimensional data. Depthwise Separable Convolution significantly reduces computational complexity by decomposing the convolution operation into depthwise convolution and pointwise convolution, while maintaining the effectiveness of feature extraction.

[0077]

[0078] Y pointwise =Y depthwise *K pointwise

[0079] Among them, X i is the i-th channel of the input feature map; K i is the i-th channel of the depth convolution kernel; K pointwise is a 1×1 convolution kernel for channel mixing; Y depthwise Represents the output feature map after depth convolution; Y pointwise Represents the output feature map after point-by-point convolution.

[0080] Depthwise convolution performs convolution operations on each input channel separately without involving interaction between channels. Point-by-point convolution mixes the output of depthwise convolution through 1×1 convolution kernel to generate the final feature map. Therefore, let the input feature map be The convolution kernel is Where H and W are the height and width of the feature map respectively, Cin is the number of input channels, C out is the number of output channels, k H and k W are the height and width of the convolution kernel.

[0081] F 传统 =H×W×C in ×k H ×k W ×C out

[0082] F 深度可分离 =H×W×C in ×k H ×k W +H×W×C in ×C out

[0083] By introducing depthwise separable convolution, the computational cost (F) is significantly reduced while the feature extraction capability of the model is not affected.

[0084] In the diagnostic model, to better capture long-term dependencies in time series data, this example introduces relative position embedding (Relative Position Embedding) into the Transformer model. Relative position embedding calculates the relative distance between any two positions in the input sequence instead of absolute position encoding, thereby better capturing long-distance dependencies.

[0085] Absolute Position Embedding (APE) directly embeds position information into the input sequence, but it is difficult to capture long-distance dependencies. Relative Position Encoding (RPE) is introduced to represent position information by calculating the relative distance between any two positions i and j, which can better capture long-distance dependencies. By introducing relative position encoding, the Transformer model can better capture long-term dependencies in time series data, thereby improving the accuracy of fault diagnosis. The calculation formula for the relative position encoding involved is as follows:

[0086]

[0087] At the same time, this embodiment also adopts the self-attention mechanism in the Transformer model, assuming that the input sequence Where L is the sequence length and D1 is the feature dimension, the attention output is calculated as:

[0088]

[0089] Through this mechanism, the model can capture the long-term dependencies between positions in the sequence, which plays an important role in analyzing the time series characteristics of furnace operation data.

[0090] To generate more targeted simulated fault data, this paper introduces a conditional generative adversarial network (cGAN). cGAN introduces conditional information into the generator and discriminator, enabling the generator to generate corresponding simulated data based on specific fault types. The conditional information can be a fault label or other fault-related features.

[0091] The cGAN model is used to generate simulated fault data augmentation datasets, which consists of a generator G and a discriminator D2. Its objective function is:

[0092]

[0093] Among them, x is the real data; y is the conditional information; z is the noise data; is the real data distribution p data The expected value of (x), p data (x) is the probability distribution of the real data sample; is the noise data distribution p z The expected value of (z), p z (z) is the probability distribution of the noise data sample.

[0094] Generator G generates simulated fault data G(z, y) based on the input noise z and conditional information y. Discriminator D is used to determine whether the input data is real data, taking the conditional information y into account. By incorporating conditional information, the generator can generate simulated data related to specific fault types, thereby enhancing the diversity and specificity of the dataset.

[0095] In practical applications, data augmentation techniques often require combining multiple methods to maximize data diversity. For example, vibration signals can be time-shifted and noise injected simultaneously, while temperature data can be randomly scaled and normalized. These operations can be randomly combined to generate a large number of new samples, thereby expanding the dataset. The steps are as follows:

[0096] (1) Time shift and noise injection

[0097] X aug (t) = X(t+Δt) + ∈(t)

[0098] (2) Random Scaling and Normalization

[0099]

[0100] Here, Δt is the time step of the random translation, ∈(t) is Gaussian noise, α is a random scaling factor, and the normalization operation ensures that the data are within the same dimensional range. Through these comprehensive data augmentation operations, the model can learn a more diverse data distribution, thereby improving its robustness and generalization ability in real-world applications.

[0101] In the process of model fusion, the local features F extracted by CNN are CNN and the long-term dependency feature F captured by Transformer Trans To perform fusion, the fusion method is concatenation:

[0102] F tuse =[F CNN ; F Trans ]

[0103] Among them, F fuse represents the fused features, which more comprehensively represent the operating status of the furnace and provide richer information for fault classification; [·;·] represents the splicing operation.

[0104] Fault classification is implemented using a fully connected layer, with fusion features The weight matrix of the fully connected layer is The bias vector is Where C1 is the number of fault categories, the classification output y is calculated as:

[0105] y=softmax(W1F fuse +b)

[0106] The output is converted into a probability distribution through the softmax function to determine the type and severity of the furnace fault.

[0107] In the process of model feedback and optimization, the loss function is used to guide the adjustment of model parameters according to the actual fault handling results and new data. Let the model prediction output be y pred , the true label is y true , then the cross entropy loss function is:

[0108]

[0109] Where c represents the category index.

[0110] By minimizing the loss function and using optimization algorithms such as gradient descent to update the model parameters, the model diagnostic performance is continuously improved.

[0111] Finally, the diagnostic module utilizes a diagnostic model to diagnose metallurgical furnace faults. The fault type, severity, and treatment recommendations output by the model must be closely aligned with the actual furnace operation and maintenance requirements. For example, for minor faults, operators may be advised to monitor and conduct regular inspections; for moderate faults, maintenance personnel should be promptly dispatched for inspection and treatment; and for severe faults, the system should be shut down immediately for repairs to prevent further damage from escalating. Furthermore, the system continuously optimizes treatment recommendations based on historical fault data and maintenance experience, providing strong support for stable furnace operation.

[0112] Newly generated fault data must be rigorously screened and labeled to ensure data quality and accuracy. Feedback data collected should include pre-fault warnings, real-time data at the time of the fault, and post-fault recovery data, allowing the model to comprehensively learn the evolution and characteristics of faults. Through continuous feedback and optimization, the system can adapt to the aging of furnace equipment, changes in operating conditions, and new fault modes, maintaining consistently high diagnostic performance.

[0113] The main fault categories include: vibration fault category, temperature fault category, pressure fault category, comprehensive fault category and other fault categories.

[0114] (1) Vibration fault type

[0115] ① Abnormal vibration amplitude: The vibration signal amplitude detected by the vibration sensor exceeds the normal range. This may be due to uneven material distribution within the furnace, resulting in an uneven load during operation and increased vibration amplitude. Alternatively, the furnace's support structure may be loose or damaged, affecting the furnace's stability during operation and causing abnormal vibration.

[0116] ② Abnormal vibration frequency: The frequency components of the vibration signal change. For example, new frequency components appear or the amplitude of existing frequency components changes significantly. This may be due to a malfunction of rotating components inside the kiln, such as the agitator or conveyor. When these components have unstable rotational speeds or unbalanced masses, specific frequency characteristics will be generated in the vibration signal.

[0117] ③ Abnormal vibration phase: The phase of the vibration signal changes abnormally. Under normal circumstances, the vibration phase of each part of the furnace has a certain regularity. When the internal structure of the furnace deforms or the relative positions of components change, the vibration phase will change abnormally. For example, a loose seal between the furnace body and the furnace cover, resulting in gas leakage, may affect the vibration phase relationship.

[0118] (2) Temperature fault

[0119] ① Overtemperature: The temperature value detected by the temperature sensor exceeds the preset safety threshold. This may be due to a malfunction in the furnace's heating system, such as a short circuit in the heating element or a malfunction in the control circuit, resulting in excessive heating power. Alternatively, the furnace's insulation may be damaged, causing heat to dissipate too quickly. To maintain the furnace temperature, the heating system must work excessively, causing the temperature to rise abnormally.

[0120] ② Temperature too low: The temperature falls below the range required for normal production. This could be due to a heating system malfunction, such as aging heating elements or insufficient burner air supply, resulting in insufficient heating power. Alternatively, excessive material input or changes in material properties could cause excessive heat absorption within the furnace, preventing the set temperature from being reached.

[0121] Uneven temperature distribution: Temperatures vary significantly across the furnace. This could be due to improper material stacking, resulting in uneven heat transfer. It could also be due to a malfunction in the furnace's heat exchange system, such as a clogged heat exchanger or poor airflow, preventing even heat distribution throughout the furnace.

[0122] (3) Pressure failure

[0123] ① Excessive pressure: The pressure value detected by the pressure sensor exceeds the design upper limit. This could be due to excessive air intake or poor exhaust flow, causing the pressure inside the furnace to gradually increase. For example, a burner's air supply valve could be faulty and remain fully open, allowing a large amount of gas to enter the furnace and burn, resulting in excessive pressure. Alternatively, the furnace's air outlet could be blocked, such as a clogged chimney or bent pipe, preventing the proper exhaust of flue gases.

[0124] ② Low pressure: The pressure value is below the normal range. This may be due to poor sealing performance of the furnace, resulting in leaks, which cause gas loss and pressure drop. It may also be due to a malfunction of the furnace fan, such as worn fan blades or insufficient motor power, which cannot provide sufficient air volume, making it impossible to maintain the normal pressure level in the furnace.

[0125] ③ Frequent pressure fluctuations: The pressure value fluctuates significantly multiple times within a short period of time. This may be due to an unstable combustion system in the furnace, such as an unstable gas supply or inefficient burner, resulting in alternately intense and weak combustion, causing frequent pressure fluctuations. It may also be due to the combustion of materials in the furnace producing a large amount of volatile substances. The sudden release of these substances causes a momentary increase in pressure, which then decreases due to combustion.

[0126] (4) Comprehensive fault type

[0127] ① Combined vibration and temperature failure: Abnormal vibration and temperature occur simultaneously. For example, if a supporting bearing in a furnace is damaged, it will not only cause the furnace's vibration to intensify, but may also cause the temperature near that part to rise. This is because the damaged bearing generates frictional heat, causing a localized temperature rise. Furthermore, the transmission of vibration may affect the distribution of materials within the furnace, further leading to an uneven temperature field within the furnace.

[0128] ② Combined vibration and pressure failure: Abnormal vibration and pressure occur simultaneously. For example, unstable gas flow inside the furnace creates vortices or impacts, which can cause both furnace wall vibration and pressure fluctuations inside the furnace. This combined failure may be caused by a malfunction in the furnace's air intake system, such as unstable valve opening in the intake line, resulting in fluctuating gas flow and causing both vibration and pressure abnormalities.

[0129] ③ Combined temperature and pressure failure: Abnormal changes in both temperature and pressure occur simultaneously. For example, a failure in a furnace's heat exchange system may cause both the temperature and pressure inside the furnace to rise or fall simultaneously. If the heat exchanger's efficiency decreases, heat inside the furnace cannot be dissipated quickly enough, causing both the temperature and pressure to gradually rise. Conversely, if a heat exchanger leaks, allowing a large amount of cooling medium to enter the furnace, this may cause both the temperature and pressure to drop simultaneously.

[0130] (5) Other fault types

[0131] ① Refractory damage: By long-term monitoring of temperature and pressure changes inside the furnace, combined with analysis of vibration signals, it is possible to determine whether the refractory material is damaged. For example, when cracks or spalling occur in the refractory material, the temperature inside the furnace may increase locally near the damaged area, and the vibration signal may become abnormal due to changes in the furnace wall structure.

[0132] ② Furnace structural deformation: By comprehensively analyzing multi-dimensional data such as the furnace's vibration, temperature, and pressure, it is possible to detect whether the furnace has experienced structural deformation. For example, if a part of the furnace is dented or twisted, this will cause changes in the vibration characteristics, temperature distribution, and pressure tolerance of that part. The model's fault diagnosis function can identify such faults.

[0133] ③ Heating element failure: A heating element failure can cause not only abnormal temperatures but also changes in pressure and vibration within the furnace. For example, a short circuit in a heating element can cause localized overheating, generating thermal stress and causing vibration within the furnace. Furthermore, sparks from a short circuit can cause a sudden surge in pressure within the furnace.

[0134] The system framework of this embodiment is as follows Figure 1As shown in the figure, the data flow and processing flow between each layer are shown in detail. From sensor data acquisition to final fault diagnosis and classification, each module works together to achieve comprehensive monitoring and intelligent diagnosis of metallurgical furnaces.

[0135] During implementation, this system can be deployed on-site at metallurgical enterprises and tightly integrated with furnace equipment. Real-time sensor data transmission and inter-module communication are achieved via industrial Ethernet or wireless communication networks. Servers can utilize high-performance computing clusters to meet the computational demands of multi-model training and real-time data processing. Furthermore, the system features a user-friendly human-computer interface, allowing operators to monitor furnace operating status in real time, view fault diagnosis results and troubleshooting recommendations, and configure and manage system parameters.

[0136] Example 2

[0137] This embodiment also provides a metallurgical furnace fault diagnosis method based on FL-CNN-Transformer-cGAN, Figure 2 It clearly depicts each step of the method process and its sequence, presenting the complete process from data acquisition, preprocessing, model training, fault diagnosis to model optimization. Each step is closely linked to ensure that the system can continuously and accurately evaluate and warn the operating status of the furnace.

[0138] Starting with real-time collection of multi-dimensional data such as vibration, temperature, and pressure from sensors outside the furnace, the data is filtered to remove high-frequency noise and normalized to the same dimension. Dimensionality reduction techniques such as PCA are then used to extract key features, reducing data dimensionality and computational complexity. Subsequently, based on the federated learning framework, each client uses local data to train CNN, Transformer, and cGAN models, and the server aggregates client model parameters to generate a global model. The preprocessed real-time data is input into the trained global model. The model uses the learned features and patterns to determine whether the furnace has a fault and the type of fault, and further classifies the severity of the fault, while providing corresponding handling recommendations. Finally, actual fault handling results and newly generated fault data are collected and fed back to the model training module to adjust and optimize model parameters and continuously improve diagnostic performance. The entire process is logically rigorous and interconnected, ensuring that the system can continuously and accurately evaluate and warn of the furnace's operating status.

[0139] The diagnostic model consists of a CNN model, a Transformer model, a cGAN model, and a fully connected layer. The CNN model is used to extract data features; the Transformer model is used to capture long-term dependencies; the cGAN model is used to generate an augmented dataset of simulated fault data; and the fully connected layer is used for fault classification. The diagnostic model is trained using a federated learning framework, with different clients training the CNN, Transformer, and cGAN models. Finally, the server aggregates the client model parameters to generate a global model, which is used as the diagnostic model.

[0140] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN, characterized by: include: Acquisition module, processing module, construction module and diagnosis module; The acquisition module is used to collect multi-dimensional data during the operation of the metallurgical furnace in real time; The processing module is used to pre-process the multi-dimensional data to obtain processed data; The construction module is used to construct a diagnostic model based on the processed data; The diagnostic module is used to complete the diagnosis of metallurgical furnace faults using the diagnostic model.

2. The metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN according to claim 1 is characterized in that: The acquisition module includes: a vibration sensor, a temperature sensor, and a pressure sensor; The multi-dimensional data collected includes vibration signals of the metallurgical furnace during operation, temperature data at different locations inside the furnace, and internal pressure data; The acquisition module is connected to the processing module via a wired or wireless manner.

3. The metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN according to claim 2 is characterized in that: The vibration signal includes: vibration amplitude, vibration frequency and vibration phase; The temperature data includes: the temperature value and the uniformity of temperature distribution; The pressure data includes: the size of the pressure value, the frequency of pressure fluctuations and the amplitude of pressure fluctuations.

4. The metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN according to claim 2 is characterized in that: The processing module uses a low-pass filter to remove high-frequency noise in the vibration signal; The processing module normalizes the temperature data and the pressure data, and reduces the dimension of the temperature data and the pressure data by using a principal component analysis technique.

5. The metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN according to claim 1 is characterized in that: The diagnostic model includes: a CNN model, a Transformer model, a cGAN model and a fully connected layer; The CNN model is used to extract data features of the processed data; The Transformer model is used to capture the long-term dependencies of the processed data; The cGAN model is used to generate a simulated fault data augmentation dataset; The fully connected layer is used for fault classification.

6. The metallurgical furnace fault diagnosis system based on FL-CNN-Transformer-cGAN according to claim 5 is characterized in that: The diagnostic model is trained using a federated learning framework, using different clients to train the CNN model, the Transformer model, and the cGAN model; Finally, the server aggregates the client model parameters to generate a global model, and the global model is used as the diagnosis model.

7. A metallurgical furnace fault diagnosis method based on FL-CNN-Transformer-cGAN, the method is applied to the system according to any one of claims 1 to 6, characterized in that the steps include: Real-time collection of multi-dimensional data during the operation of metallurgical furnaces; Preprocessing the multi-dimensional data to obtain processed data; constructing a diagnostic model based on the processed data; The diagnosis model is used to diagnose metallurgical furnace faults.

8. The metallurgical furnace fault diagnosis method based on FL-CNN-Transformer-cGAN according to claim 7 is characterized in that: The diagnostic model includes: a CNN model, a Transformer model, a cGAN model and a fully connected layer; The CNN model is used to extract data features of the processed data; The Transformer model is used to capture the long-term dependencies of the processed data; The cGAN model is used to generate a simulated fault data augmentation dataset; The fully connected layer is used for fault classification.

9. The metallurgical furnace fault diagnosis method based on FL-CNN-Transformer-cGAN according to claim 7 is characterized in that: The diagnostic model is trained using a federated learning framework, using different clients to train the CNN model, the Transformer model, and the cGAN model; Finally, the server aggregates the client model parameters to generate a global model, and the global model is used as the diagnosis model.

Images