Self-adaptive cooling and visual hearth thermal imaging intelligent monitoring system

Abstract

The invention discloses a self-adaptive cooling and visual hearth thermal imaging intelligent monitoring system. The system is characterized by comprising a camera lens tube system, a cooling gas source and a flow regulating valve, the camera lens tube system comprises an infrared thermal imaging camera, a lens tube cooling protective cover, a cooling air series pipeline, a lens group and a temperature transmitter; the cooling air source is connected to the cooling air series-connection pipeline through a cooling air pipeline, and the temperature transmitter is installed in the mirror tube cooling protective cover; the system further comprises a controller, when the camera lens tube system extends into the furnace for monitoring, the temperature transmitter collects the temperature in the lens tube cooling protective cover, and the controller adjusts the opening degree of the flow adjusting valve in real time according to the temperature fed back by the temperature transmitter, so that the cooling air intensity is controlled. The working temperature is monitored through the temperature sensor, data are transmitted to the controller, the controller adjusts the rotating speed of the draught fan or the opening degree of the air valve according to a preset cooling strategy, and therefore dynamic adjustment of the cooling air intensity is achieved.

Description

Technical Field

[0001] The present invention relates to the field of furnace detection, and particularly to an intelligent monitoring system for adaptive cooling and visual furnace thermal imaging. Background Art

[0002] Among the already disclosed patents, there are a furnace coking detection system (CN220932801 U) and a furnace flame monitoring device (CN215982664U). These patents disclose different types of traditional furnace monitoring devices, which can provide thermal imaging monitoring functions to a certain extent. However, in terms of system cooling control, only a cooling gas pressure sensor or a temperature sensor is provided. When the temperature is too high or the cooling gas pressure is too low, the camera lens tube is withdrawn in a timely manner to achieve the protection function. However, this cooling protection method cannot be automatically adjusted and is often affected by unstable air pressure, resulting in shutdown and withdrawal problems, and cannot ensure long-term stable operation. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent monitoring system for adaptive cooling and visual furnace thermal imaging.

[0004] To achieve the above purpose, the present invention provides the following technical solutions:

[0005] An intelligent monitoring system for adaptive cooling and visual furnace thermal imaging, characterized in that it includes a camera lens tube system, a cooling gas source, and a flow regulating valve;

[0006] The camera lens tube system includes an infrared thermal imaging camera, a lens tube cooling protective cover, a cooling air series pipeline, a lens group, and a temperature transmitter;

[0007] The cooling gas source is connected to the cooling air series pipeline through a cooling gas pipeline, and the temperature transmitter is installed inside the lens tube cooling protective cover;

[0008] It further includes a controller. When the camera lens tube system extends into the furnace for monitoring, the temperature transmitter collects the temperature inside the lens tube cooling protective cover, and the controller adjusts the opening of the flow regulating valve in real time according to the temperature fed back by the temperature transmitter, so as to control the cooling air intensity.

[0009] Preferably, it further includes a gas source pressure transmitter and a flow sensor, and the gas source pressure transmitter and the flow sensor are installed in the cooling gas pipeline between the cooling gas source and the lens tube cooling protective cover.

[0010] Preferably, it further includes an advancing and retreating mechanism for driving the camera lens tube system to enter and exit the furnace. The advancing and retreating mechanism includes a driving motor, a conveying chain, and a conveying hoist.

[0011] Preferably, the camera tube system collects flame images through an infrared thermal imaging camera. The system presets a flame model, analyzes the flame images in real time, and judges the flame state for coking detection.

[0012] Preferably, the flame model uses a multi-layer perceptron to extract the characteristics of the burner operating conditions data, uses Vision Transformer to extract the global characteristics of the image, uses Canny edge detection to segment the flame area, and then uses ViT to extract the local characteristics of the image; after splicing the above characteristics to form a multi-modal fusion feature, it is input into a long short-term memory network for time series feature analysis, and finally a Softmax layer is used for classification.

[0013] Preferably, the model uses labeled data to jointly train the MLP model, ViT model and classifier. During the training process, a cross-entropy loss function is adopted, and the model parameters are updated through the backpropagation algorithm. The training data is divided into a training set, a validation set and a test set to ensure the generalization ability of the model. During the training process, data augmentation techniques such as random rotation, scaling, flipping, etc. are also adopted to increase the diversity of the data;

[0014] The data input into the flame model is continuously acquired burner operating conditions data and flame morphology images.

[0015] Preferably, the training process of the flame model includes the following steps:

[0016] Data acquisition: Flame image data: A high-resolution industrial camera is used to collect flame images at a frame rate of not less than 30fps, covering the visible light and near-infrared spectral ranges to ensure clear capture of the details of the flame morphology. Each frame of image is marked with a timestamp and the flame state label is synchronously marked, namely normal, incomplete combustion, flame disappearance, coking occlusion;

[0017] Operating condition parameters: 7 types of burner operating condition parameters are collected in real time through industrial-grade sensors, the data sampling frequency is 1Hz, and linear interpolation is performed according to the frame rate of the image to align with the image data timestamp;

[0018] Feature extraction: The input operating condition data is used by the MLP module for feature extraction. The MLP module is a 3-layer fully connected network, and the ReLU activation function is used in the hidden layer to output a 32-dimensional operating condition feature vector;

[0019] Image global feature extraction, input image data, and use ViT-Global for feature extraction;

[0020] Image local feature extraction: Use Canny edge detection to extract the flame area, crop it into a 112×112 local image, and use ViT-Local for local feature extraction.

[0021] Preferably, the flame model further includes feature fusion and temporal modeling steps:

[0022] Feature splicing method: splice the working condition features extracted by MLP, the ViT-Global global image features, and the ViT-Local local image features into 1184-dimensional fusion features;

[0023] Use a long short-term memory (LSTM) network for feature modeling and pattern classification.

[0024] Compared with the prior art, the beneficial effects of the present invention are: the working temperature is monitored by a temperature sensor and the data is transmitted to the controller, and the controller adjusts the fan speed or the opening degree of the air valve through a preset cooling strategy, so as to realize the dynamic adjustment of the cooling air intensity. Through real-time feedback, the system can ensure that the temperature is within a safe range, thus effectively protecting the stability and service life of the furnace monitoring system. Description of the Drawings

[0025] Figure 1 It is a schematic structural diagram of the camera tube system of the present invention.

[0026] Figure 2 It is a schematic diagram of the state of the camera tube system of the present invention when it is sent into the furnace.

[0027] Figure 3 It is a schematic structural diagram of the flame model of the present invention. Detailed Embodiments

[0028] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0029] Embodiment: Please refer to Figure 1 , the present invention provides an adaptive cooling and visual furnace thermal imaging intelligent monitoring system, including a camera tube system, a cooling gas source, and a flow regulating valve;

[0030] The camera tube system includes an infrared thermal imaging camera, a lens tube cooling protective cover 2.2, a cooling air series pipeline 2.3, a lens group 2.4, a temperature transmitter 2.5, a gas source pressure transmitter, and a flow sensor;

[0031] The cooling gas source is connected to the cooling air series pipeline through a cooling gas pipeline, and the temperature transmitter is installed inside the lens tube cooling protective cover;

[0032] It also includes a controller. When the camera tube system extends into the furnace for monitoring, the temperature transmitter collects the temperature inside the mirror tube cooling protective cover. The controller adjusts the opening degree of the flow regulating valve in real time according to the temperature feedback by the temperature transmitter, so as to control the cooling air intensity.

[0033] It also includes a retracting and extending mechanism. The retracting and extending mechanism 1 consists of a housing 1.1, a driving motor 1.2, a conveying chain 1.3, and a conveying hoist 1.4.

[0034] After the system starts, the gas source pressure transmitter begins to collect the cooling air pressure, and the high-precision temperature transmitter 2.5 begins to collect the temperature inside the mirror tube cooling protective cover. The above information is sent to the on-site controller 4 and then uploaded to the remote server for processing. The data acquisition module system judges whether each parameter is within the normal range. After meeting the start-up conditions, the flow regulating valve opens (opening degree ≤ 10%), and the adaptive cooling unit starts to operate. The driving motor 1.2 of the retracting and extending actuator starts and makes a linear motion through the conveying chain 1.3 and the conveying hoist 1.4 to extend the camera tube system into the flue or furnace (such as Figure 1 and Figure 2 ), and the temperature transmitter 2.5 begins to monitor the working temperature of the camera tube in real time.

[0035] The controller determines the cooling air intensity according to the temperature data and the preset cooling strategy. Different temperature thresholds are set in the system, and the corresponding cooling air intensity is preset for each temperature range. For example:

[0036] Low temperature area: When the temperature is relatively low, the opening degree of the flow regulating valve is automatically adjusted in the range of 11% - 30%. While ensuring that the temperature inside the mirror tube cooling protective cover ≤ the set value (such as 40°C), the cooling air intensity is kept at the lowest level to save energy.

[0037] Medium temperature area: When the temperature continuously rises to 50°C, the opening degree of the flow regulating valve increases and is automatically adjusted in the range of 31% - 60%. If the temperature drops below 40°C, the opening degree of the flow regulating valve is adjusted down to the low temperature control area accordingly.

[0038] High temperature area: When the temperature continues to rise to 60°C and enters the high temperature range, the opening degree of the flow regulating valve increases and is automatically adjusted in the range of 61% - 100%. The cooling air intensity is the maximum to protect the camera lens.

[0039] If the temperature drops to the previous temperature range, the valve opening degree is adjusted accordingly to save energy. If the valve opening degree is 100% and the temperature continues to rise to 70°C, the system will alarm immediately, and the camera tube will be forced to withdraw to protect the lens and the camera.

[0040] The system adopts closed-loop feedback control. By adjusting the opening of the flow regulating valve through real-time temperature feedback, it controls the wind speed to achieve a stable adaptive effect. The system is built-in with a gas source pressure transmitter and a flow sensor to monitor the flow rate and pressure of the cooling air, ensuring that the air volume and pressure reach the expected values. If it detects insufficient cooling air or the temperature exceeds the preset upper limit, the controller will send an alarm signal or automatically take protective measures (such as shutting down the equipment or forcibly increasing the cooling air).

[0041] By importing a large amount of flame data obtained from laboratory tests and industrial sites, including normal flame morphology data and abnormal (including incomplete combustion, flame disappearance, flame occlusion caused by coking, etc.) flame data. These data are preprocessed, including steps such as denoising, normalization, and data augmentation. During the operation of the furnace, the flame model infers from the real-time collected images and sensor data, analyzes the flame morphology in real time, and performs abnormal detections such as flame state judgment and coking detection.

[0042] The flame model is a multi-modal deep learning model. The input data includes burner operating condition parameters and flame morphology images, and the output data is the labeled flame state result corresponding to the input data;

[0043] Collect burner flame image data and burner operating condition data. The flame image data is collected by a high-resolution industrial camera to ensure the clarity and details of the images. The burner operating condition data is collected in real time by sensors to ensure the timeliness and accuracy of the data.

[0044] The structure of the flame model is as follows: Use a multi-layer perceptron (MLP) to extract the features of the burner operating condition data, use a Vision Transformer (ViT) to extract the global features of the image, use Canny edge detection and segment the flame area, and then use ViT to extract the local features of the image; After splicing the above features to form a multi-modal fusion feature, input it into a long short-term memory network (LSTM) for temporal feature analysis, and finally use a Softmax layer for classification;

[0045] The flame model uses labeled data to jointly train the MLP model, ViT model, and classifier. During the training process, a cross-entropy loss function is adopted, and the model parameters are updated through the backpropagation algorithm. The training data is divided into a training set, a validation set, and a test set to ensure the generalization ability of the model. Data augmentation techniques such as random rotation, scaling, and flipping are also adopted during the training process to increase the diversity of the data.

[0046] The input data of the flame model is continuously acquired burner operating condition data and flame morphology images.

[0047] The method has the advantages of high detection accuracy, strong robustness, and adaptability to complex working conditions.

[0048] The following are the detailed implementation steps:

[0049] 1. Data acquisition

[0050] Flame image data: A high-resolution industrial camera (resolution ≥ 2048×1536) is used to collect flame images at a frame rate of not less than 30fps, covering the visible and near-infrared spectral ranges to ensure clear capture of flame morphology details (such as brightness distribution and edge contours). Each frame of the image is marked with a timestamp and the flame status label is synchronously annotated, namely normal, incomplete combustion, flame disappearance, and coking occlusion.

[0051] Operating condition parameters: Seven types of burner operating condition parameters (as shown in Table 1) are collected in real time through industrial-grade sensors. The data sampling frequency is 1Hz, and linear interpolation is performed according to the frame rate of the image to align with the image data timestamp.

[0052] Table 1 Collected burner operating condition data

[0053]

[0054]

[0055] 2. Data preprocessing

[0056] 2.1 Boiler operating condition data acquisition and preprocessing

[0057] The operating condition parameters (Table 1) are standardized using Z-score, and the formula is as follows:

[0058]

[0059] where μ is the mean of the training set and σ is the standard deviation.

[0060] 2.2 Image data preprocessing

[0061] Gaussian filtering (kernel size 5×5, σ = 1.5) is applied to eliminate image noise, and histogram equalization is combined to enhance the contrast of the flame area.

[0062] Data augmentation includes random horizontal / vertical flipping (probability 0.5), rotation (angle range ±15°), scaling (scale 0.9 - 1.1), and brightness adjustment (±20%).

[0063] 3. Multimodal deep learning model

[0064] 3.1 Feature extraction module

[0065] 3.1.1 Operating condition parameter feature extraction (MLP module)

[0066] The input and network structure form of the MLP module are as follows:

[0067] Input: The working condition parameters after standardization, i.e., the 7-dimensional working condition parameters described in Table 1.

[0068] Structure: A 3-layer fully connected network (input layer 7 → hidden layer 128 → hidden layer 64 → output layer 32). The ReLU activation function is used in the hidden layer, and a 32-dimensional working condition feature vector is output.

[0069] 3.1.2 Image global feature extraction (ViT-Global):

[0070] To extract the overall features of the images obtained by the industrial camera, the input and network structure of ViT-Global are as follows:

[0071] Input: The RGB image is scaled to a resolution of 224×224 and divided into blocks of 16×16 pixels (a total of 196 blocks).

[0072] Structure: Vision Transformer (ViT-Base), which contains 12 layers of Transformer encoders, with a multi-head attention mechanism (12 heads) in each layer. An output feature map of 196×768 dimensions is obtained, and a 768-dimensional global feature vector is obtained through global average pooling.

[0073] 3.1.3 Image local feature extraction (ViT-Local):

[0074] To further extract the detailed local features of the flame in the images obtained by the industrial camera, the input and network structure of ViT-Local are as follows:

[0075] Input: The flame region is extracted using Canny edge detection and cropped into a local image of 112×112.

[0076] Structure: A lightweight ViT (ViT-Small) with a block size of 8×8, 6 layers of Transformer encoders. An output feature map of 64×384 dimensions is obtained, and a 384-dimensional local feature vector is obtained through max pooling.

[0077] 3.2 Multimodal feature fusion and temporal modeling

[0078] To achieve the fusion of the features extracted from the working condition data and the image features, multimodal feature fusion is required:

[0079] Feature splicing method: The working condition features (32 dimensions) extracted by the MLP, the ViT-Global global image features (768 dimensions), and the ViT-Local local image features (384 dimensions) are spliced into a 1184-dimensional fusion feature.

[0080] Considering that both the working condition data and the image data have obvious temporal characteristics, a long short-term memory (LSTM) network is adopted for feature modeling and pattern classification:

[0081] Input: A continuous 8-second fusion feature sequence (i.e., the time step of the data is 8, and the feature dimension is 1184).

[0082] Structure: A two-layer LSTM (the number of hidden units is 256), and the output of the last hidden state (256 dimensions) is used as the temporal feature.

[0083] Classifier: A multi-layer fully connected layer (256→128→4) combined with Softmax outputs the probabilities of 4 categories (normal, incomplete combustion, flame disappearance, coking occlusion).

[0084] This embodiment fuses the working condition parameters of the burner and the multi-modal data of the flame morphology images, captures the physical state and visual features of the boiler simultaneously, and solves the problem of the information limitation of single-modal data; at the same time, it coordinates the global and local features, and improves the sensitivity to local anomalies synchronously while increasing the receptive field;

[0085] The LSTM is used to perform temporal analysis on the continuous fusion features, effectively capturing the dynamic change rules of the flame (such as the flame flicker frequency and the gradual change process of combustion efficiency), and avoiding the misjudgment risk of single-frame detection.

[0086] For those skilled in the art, it is obvious that the present invention is not limited to the details of the above-described exemplary embodiments, and can be implemented in other specific forms without departing from the spirit or basic characteristics of the present invention. Therefore, in any regard, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is defined by the appended claims rather than the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the claims are intended to be included in the present invention. Any reference signs in the claims should not be regarded as limiting the claimed rights.

Claims

1. An adaptive cooling and visual furnace thermal imaging intelligent monitoring system, characterized by: It includes a camera tube system, a cooling gas source and a flow regulating valve; The camera tube system includes an infrared thermal imaging camera, a tube cooling shield, a cooling air series duct, a lens group, and a temperature transmitter; The cooling air source is connected to the cooling air series pipe through the cooling air pipe, and the temperature transmitter is installed in the mirror tube cooling protective cover; It also includes a controller. When the camera tube system is extended into the furnace for monitoring, the temperature transmitter collects the temperature inside the tube cooling shield. The controller adjusts the opening of the flow regulating valve in real time according to the temperature feedback from the temperature transmitter, thereby controlling the cooling wind intensity.

2. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 3 is characterized by: It also includes an air source pressure transmitter and a flow sensor. The air source pressure transmitter and the flow sensor are installed in a cooling air pipeline between a cooling air source and a mirror tube cooling protective cover.

3. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 1 is characterized by: It also includes an advance and retreat mechanism, which is used to drive the camera tube system to enter and exit the furnace. The advance and retreat mechanism includes a driving motor, a conveying chain and a conveying hoist.

4. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 3 is characterized by: The camera tube system collects flame images through an infrared thermal imaging camera. The system presets a flame model, analyzes the flame images in real time, and determines the flame status for coking detection.

5. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 4 is characterized by: The flame model uses a multi-layer perceptron to extract the features of the burner operating data, uses Vision Transformer to extract the global features of the image, uses Canny edge detection and segmentation of the flame area, and then uses ViT to extract the local features of the image; the above features are spliced ​​to form a multimodal fusion feature, which is input into the long short-term memory network for temporal feature analysis, and finally classified using the Softmax layer.

6. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 3 is characterized by: The model uses labeled data to jointly train the MLP model, ViT model and classifier. The cross entropy loss function is used in the training process, and the model parameters are updated through the back propagation algorithm. The training data is divided into training set, validation set and test set to ensure the generalization ability of the model. Data enhancement techniques such as random rotation, scaling, flipping, etc. are also used in the training process to increase the diversity of data. The input data of the flame model are the continuously acquired burner operating data and flame morphology images.

7. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 6 is characterized by: The flame model training process includes the following steps: Data collection: Flame image data: A high-resolution industrial camera is used to collect flame images at a frame rate of no less than 30fps, covering the visible light and near-infrared spectrum range to ensure clear capture of flame morphology details. Each frame of the image is marked with a timestamp and the flame status label is synchronously annotated, i.e. normal, incomplete combustion, flame disappearance, and coking occlusion; Working condition parameters: 7 types of burner working condition parameters are collected in real time through industrial-grade sensors. The data sampling frequency is 1Hz, and linear interpolation is performed according to the frame rate of the image, and aligned with the image data timestamp; Feature extraction: The input working condition data is extracted using the MLP module. The MLP module is a 3-layer fully connected network. The hidden layer uses the ReLU activation function and outputs a 32-dimensional working condition feature vector. Image global feature extraction, input image data, and use ViT-Global to extract features; Image local feature extraction: Canny edge detection is used to extract the flame area, which is cropped into a 112×112 local image, and ViT-Local is used to extract local features.

8. The adaptive cooling and visualized furnace thermal imaging intelligent monitoring system according to claim 7 is characterized by: The flame model also includes feature fusion and timing modeling steps: Feature splicing method: The working condition features extracted by MLP, ViT-Global global image features and ViT-Local local image features are spliced ​​into 1184-dimensional fusion features; The long short-term memory (LSTM) network is used for feature modeling and pattern classification.

Images