WO2024060285A1 - 一种基于红外测温的高炉软十字测温方法 - Google Patents

一种基于红外测温的高炉软十字测温方法 Download PDF

Info

Publication number
WO2024060285A1
WO2024060285A1 PCT/CN2022/122616 CN2022122616W WO2024060285A1 WO 2024060285 A1 WO2024060285 A1 WO 2024060285A1 CN 2022122616 W CN2022122616 W CN 2022122616W WO 2024060285 A1 WO2024060285 A1 WO 2024060285A1
Authority
WO
WIPO (PCT)
Prior art keywords
temperature measurement
temperature
cross
blast furnace
infrared
Prior art date
Application number
PCT/CN2022/122616
Other languages
English (en)
French (fr)
Inventor
蔡炜
严晗
林子恒
Original Assignee
中冶南方工程技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 中冶南方工程技术有限公司 filed Critical 中冶南方工程技术有限公司
Publication of WO2024060285A1 publication Critical patent/WO2024060285A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00—Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/0044—Furnaces, ovens, kilns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the invention relates to the field of blast furnace ironmaking in the metallurgical industry, and in particular to a blast furnace soft cross temperature measurement method based on infrared temperature measurement.
  • the cross temperature measuring device has been widely used in bellless distribution blast furnaces. It can continuously and accurately measure the temperature distribution of the gas flow in the radial direction of the furnace throat. Since the gas flow is strong in places with high temperatures, the cross temperature measuring device can effectively monitor the top of the blast furnace. Distribution of gas flow.
  • the cross temperature measuring device is located at the blast furnace throat, and four temperature measuring arms are installed in four directions on the circumferential surface of the furnace throat. Each temperature measuring wall is distributed with different temperature sensors (such as thermocouples). ), with a total of 17-21 temperature measurement sensors. These temperature measurement points can provide real-time temperature data and can more comprehensively reflect the distribution of gas flow in the circumferential direction of the furnace throat.
  • the cross temperature measuring device is installed on the top of the sealed blast furnace barrel, and the operating environment is extremely harsh. Therefore, the sensor of the cross temperature measuring device is easily damaged, and maintenance cannot be carried out in time after damage. The sensor must be replaced after a large period of blast furnace maintenance.
  • the non-contact infrared thermal imaging temperature measurement method was developed in the 1990s and has been widely used in blast furnace top temperature measurement, which can observe the spatial temperature field distribution in the blast furnace. Since the infrared focal plane array of thermal imaging sensing elements does not need to be in contact with the spatial target in the furnace, it will not affect the temperature field distribution and sensor life. It uses infrared thermal radiation to measure temperature. It has the advantages of fast response, long sensor life, non-consumable, high measurement temperature, and can achieve real-time continuous measurement.
  • the temperature measurement result has a great relationship with the target object's emissivity.
  • the emissivity of the infrared detector is set to 0.97, the temperature result measured by the infrared detector is much lower than the cross temperature measurement result. , so the operator cannot directly replace the cross temperature measurement with the temperature measurement results of the original infrared image, and use the existing cross temperature measurement temperature model to perform corresponding blast furnace operations and decisions.
  • the purpose of the present invention is to combine the furnace top infrared temperature measurement technology with the cross temperature measurement technology without installing a traditional cross temperature measurement device, and convert the temperature measurement results of the original infrared image into a set cross
  • the temperature value of the temperature measurement point is measured, and the existing cross temperature measurement temperature model is used to perform corresponding blast furnace operations and decisions.
  • the present invention provides a blast furnace soft cross temperature measurement method based on infrared temperature measurement, which includes the following steps:
  • Step S10 set up an infrared temperature measurement device and a cross temperature measurement device at the same time on the blast furnace; determine the virtual cross temperature measurement point on the infrared image according to the position mapping relationship between the pixel points of the infrared image and the cross temperature measurement point; and determine the virtual cross temperature measurement point based on the virtual cross temperature measurement point.
  • the temperature value of the temperature point and the temperature value of the cross temperature measurement point are used to train the temperature relationship model between the virtual cross temperature measurement point and the real cross temperature measurement point;
  • Step S20 only install an infrared temperature measurement device on the blast furnace; set the position of the cross temperature measurement point, and determine the virtual cross temperature measurement point on the infrared image based on the positional mapping relationship between the pixels of the infrared image and the cross temperature measurement point; from The infrared image extracts the temperature value of the virtual cross temperature measurement point, and then outputs the predicted temperature value of the cross temperature measurement point through the temperature relationship model.
  • step S10 includes:
  • Step S101 Install a cross temperature measuring device and an infrared temperature measuring device on the blast furnace at the same time;
  • Step S102 Establish the plane coordinate system of the blast furnace according to the installation plane of the cross temperature measuring device in the furnace;
  • Step S103 Establish a perspective transformation matrix based on the installation position of the infrared temperature measurement device on the furnace top, and correct the infrared image I1 to a bird's-eye temperature image I2 located in the blast furnace plane coordinate system;
  • Step S104 According to the installation position of the cross temperature measuring device in the furnace, find the corresponding pixel coordinates in the bird's-eye view temperature image I2. Through these pixel coordinates, find the corresponding pixel coordinates in the infrared image I1, and establish a virtual cross temperature measurement. point;
  • Step S105 Read the temperature values of the virtual cross temperature measurement points to form a first temperature sequence
  • Step S106 Read the temperature value of the cross temperature measurement point through the cross temperature measurement device to form a second temperature sequence
  • Step S107 Using the first temperature sequence as input and the second temperature sequence as output, train a temperature relationship model between the virtual cross temperature measurement point and the real cross temperature measurement point.
  • step S10 the plane coordinate system of the blast furnace is set so that the origin of the coordinates of the top of the blast furnace is located at the center of the blast furnace on the horizontal plane where the temperature measurement point of the cross temperature measurement device is located.
  • step S103 includes:
  • the perspective transformation matrix P1 from the blast furnace coordinate system to the infrared image I1 coordinate system is calculated, and based on the perspective transformation matrix, the bird's-eye temperature image I2 converted from the original infrared image I1 to the horizontal coordinate system of the blast furnace is obtained.
  • step S107 the temperature relationship model is based on the MLP neural network model.
  • step S20 includes:
  • Step S201 Install an infrared temperature measurement device on the top of the blast furnace to obtain an infrared image of the furnace top;
  • Step S202 Establish a perspective transformation matrix based on the installation position of the infrared temperature measurement device on the furnace top, and correct the infrared image I3 to a bird's-eye temperature image I4 located in the blast furnace plane coordinate system;
  • Step S203 Set the position of the cross temperature measurement point
  • Step S204 Find the pixel coordinates corresponding to the cross temperature measurement point from the bird's-eye temperature image I4, and then establish a virtual cross temperature measurement point in the infrared image I3 through the perspective transformation matrix;
  • Step S205 Extract the temperature values of the virtual cross temperature measurement points to form a third temperature sequence
  • Step S206 Input the third temperature sequence into the trained temperature relationship model of virtual cross temperature measurement points and real cross temperature measurement points, and output a fourth temperature sequence composed of predicted temperature values of each cross temperature measurement point.
  • the original infrared image is obtained through the infrared temperature measurement device, and the temperature measurement results of the original infrared image are converted into the temperature value of the cross temperature measurement point, and the existing cross temperature measurement temperature is used.
  • the model performs corresponding blast furnace operations and decisions.
  • Figure 1 is a flow chart of the blast furnace soft cross temperature measurement method based on infrared temperature measurement according to the present invention
  • Figure 2 is a position map of the infrared image and bird's-eye temperature image of the present invention
  • Figure 3 is a neural network model related to the present invention.
  • this application proposes a blast furnace soft cross temperature measurement method based on infrared temperature measurement, which includes the following steps:
  • Step S10 simultaneously setting an infrared temperature measuring device and a cross temperature measuring device on the blast furnace; determining a virtual cross temperature measuring point on the infrared image according to a position mapping relationship between pixel points of the infrared image and the cross temperature measuring point; and training a temperature relationship model between the virtual cross temperature measuring point and the real cross temperature measuring point according to the temperature value of the virtual cross temperature measuring point and the temperature value of the cross temperature measuring point.
  • Step S20 only install an infrared temperature measurement device on the blast furnace; set the position of the cross temperature measurement point, and determine the position of the virtual cross temperature measurement point on the infrared image based on the positional mapping relationship between the pixels of the infrared image and the cross temperature measurement point ; Extract the temperature value of the virtual cross temperature measurement point from the infrared image, and then output the predicted temperature value of the cross temperature measurement point through the temperature relationship model.
  • step S10 includes:
  • the temperature value of the virtual cross temperature measurement point is read to form a first temperature sequence; the temperature value of the cross temperature measurement point is read through the cross temperature measurement device to form a second temperature sequence.
  • the infrared temperature measurement camera cannot be installed on the center line of the blast furnace roof and takes pictures completely parallel to the horizontal plane, it is installed on the side wall of the blast furnace. It is at a certain angle with the horizontal plane, so the concentric circle pattern of the cross temperature measurement is not a circle on the infrared image, but a pattern of concentric circles after perspective transformation, as shown in Figure 2. Therefore, the first step requires solving the perspective transformation matrix P from the actual coordinates of the blast furnace to the infrared image coordinates.
  • the coordinates on can be calculated by manual measurement or image algorithm to calculate the upper, lower, left and right boundary points to obtain A'(XA,YA),B'(XB,YB),C'(XC,YC),D'(XD,YD), here Take the point in the upper left corner of the image as the origin.
  • the perspective transformation matrix P1 from the blast furnace coordinate system to the infrared image I1 coordinate system can be calculated, and based on the perspective transformation matrix, the bird's-eye temperature image I2 converted from the original infrared image I1 to the horizontal coordinate system of the blast furnace is obtained.
  • a virtual cross temperature measurement solution can be designed, such as setting a temperature measurement point every 0.8 meters in the four directions of 0°, 90°, 180°, and 270°, and setting 5 points in each direction.
  • the coordinates of the temperature measurement point are:
  • This virtual cross temperature measurement scheme can adopt the original cross temperature measurement scheme, thereby facilitating the use of the original temperature model for blast furnace operation and decision-making. It is also possible to set up new cross temperature measurement schemes and train new temperature models to support blast furnace operations and decision-making.
  • the input layer of the network is the result of infrared temperature measurement.
  • the output layer is the fitted temperature measurement result of the corresponding cross temperature measurement point.
  • the network contains 2 hidden layers, each hidden layer contains 10 nodes, as shown in Figure 3.
  • the following temperature relationship model training scheme for virtual cross temperature measurement points and real cross temperature measurement points is given: 20 cross temperature measurement points and their corresponding infrared temperature measurement results are sampled at intervals of 2 hours. , they form a sample data.
  • This implementation plan samples a total of 12,000 sets of sample data for 50 days to perform cross-validation and solve the temperature correction network.
  • the loss function uses the mean squared error (Mean Squared Error), which is defined as:
  • M is the number of samples in a certain sample set, which includes training set, verification set and test set;
  • T(k) and T S (k) are respectively the true cross temperature measurement value measured in the kth group and the temperature measurement value output by the network.
  • the training is terminated after the error change of the validation set approaches 0, and the weights and bias parameters are saved to the model file Model.
  • neural network models such as MLP (multilayer perceptron) can be used for training.
  • MLP multi-layer perceptron
  • the MLP multi-layer perceptron is a forward-structured artificial neural network ANN that maps a set of input vectors to a set of output vectors.
  • MLP can be viewed as a directed graph consisting of multiple layers of nodes, with each layer fully connected to the next layer. Except for the input node, each node is a neuron with a nonlinear activation function.
  • step 20 includes:
  • Step S201 Install an infrared temperature measurement device on the top of the blast furnace
  • Step S202 according to the installation position of the infrared temperature measuring device on the furnace roof, a perspective transformation matrix is established to correct the infrared image I3 into a bird's-eye view temperature image I4 located in the blast furnace plane coordinate system;
  • Step S203 Set the position of the cross temperature measurement point
  • Step S204 Find the pixel coordinates corresponding to the cross temperature measurement point from the bird's-eye temperature image I4, and then establish a virtual cross temperature measurement point in the infrared image I3 through the perspective transformation matrix;
  • Step S205 Extract the temperature values of the virtual cross temperature measurement points to form a third temperature sequence
  • Step S206 Input the third temperature sequence into the trained temperature relationship model of virtual cross temperature measurement points and real cross temperature measurement points, and output a fourth temperature sequence composed of predicted temperature values of each cross temperature measurement point.
  • the fourth temperature sequence is the predicted temperature value of the cross temperature measurement point. Based on these predicted temperature values, the existing cross temperature measurement temperature model can be used to perform corresponding blast furnace operations and decisions.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Radiation Pyrometers (AREA)

Abstract

本发明涉及高炉冶金领域，公开了一种基于红外测温的高炉软十字测温方法。所述方法包括：首先，在高炉上安装十字测温装置和红外测温装置，根据红外图像的像素点和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点；并根据虚拟十字测温点的温度值和十字测温点的温度值，训练虚拟十字测温点和真实十字测温点的温度关系模型；然后在仅安装了红外测温的高炉上使用该模型来计算出十字测温点的预测温度值。采用该方法，无需安装传统的十字测温装置，将原始红外图像的测温结果转换为十字测温点的温度值，并利用现有的十字测温温度模型执行相应的高炉操作和决策。

Description

一种基于红外测温的高炉软十字测温方法 技术领域
本发明涉及冶金行业高炉炼铁领域，尤其是涉及一种基于红外测温的高炉软十字测温方法。
背景技术
十字测温装置在无钟布料高炉得到了广泛应用，它能连续准确地测出炉喉径向的煤气流温度分布.由于温度高的地方煤气流旺盛，因而十字测温装置可以有效监测高炉炉顶煤气流的分布状况.十字测温装置位于高炉炉喉位置，并在炉喉圆周面上的四个方向安装四个测温臂，每个测温壁分布有不等的温度传感器(例如热电偶)，共有17—21个测温传感器.这些测温点能够提供实时温度数据，可比较全面地反映煤气流在炉喉圆周方向上的分布。十字测温装置安装在密闭的高炉筒体顶部，运行环境极为恶劣。因此十字测温装置传感器容易损坏，损坏后不能及时进行维护，必须等到大周期的高炉检修才能进行传感器的更换。
非接触式红外热成像测温方法在20世纪90年代发展起来，目前已经广泛应用于高炉炉顶测温，能观察高炉炉内空间温度场分布。由于热成像传感元件红外焦平面阵列不需要与炉内空间目标进行接触，故不会对温度场分布及传感器寿命产生影响。它利用红外热辐射进行测温，具有响应快、传感器寿命长、非消耗性、测量温度高，可实现实时连续测量的优点。
通过红外辐射测温原理可知测温结果与目标物体辐射率有很大关系，通过实测数据发现将红外探测器的辐射率设置为0.97时，红外探测器测量的温度结果远低于十字测温结果，因此操作员不能直接以原始红外图像的测温结果来代替十字测温，并利用现有的十字测温的温度模型执行相应的高炉操作和决策。
发明内容
有鉴于现有技术的上述缺陷，本发明的目的是将炉顶红外测温技术与十字测温技术结合，无需安装传统的十字测温装置，将原始红外图像的测温结果转换为设定十字测温点的温度值，并利用现有的十字测温温度模型执行相应的高炉操作和决策。
为实现上述目的，本发明提供了一种基于红外测温的高炉软十字测温方法，包括以下步骤：
步骤S10，在高炉上同时设置红外测温装置和十字测温装置；根据红外图像的像素点和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点；并根据虚拟十字测温点的温度值和十字测温点的温度值，训练虚拟十字测温点和真实十字测温点的温度关系模型；
步骤S20，在高炉上仅设置红外测温装置；设定十字测温点的位置，根据红外图像的像素点和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点；从红外图像提取虚拟十字测温点的温度值，再通过温度关系模型，输出十字测温点的预测温度值。
进一步的，所述步骤S10，包括：
步骤S101：在高炉上同时安装十字测温装置和红外测温装置；
步骤S102：根据十字测温装置在炉内的安装平面，建立高炉的平面坐标系；
步骤S103：根据红外测温装置在炉顶的安装位置，建立透视变换矩阵，将红外图像I1校正为位于高炉平面坐标系的鸟瞰温度图像I2；
步骤S104：根据十字测温装置在炉内的安装位置，在鸟瞰温度图像I2中找到对应的像素点坐标，通过这些像素点坐标，在红外图像I1找到对应的像素点坐标，建立虚拟十字测温点；
步骤S105：读取虚拟十字测温点的温度值，形成第一温度序列；
步骤S106：通过十字测温装置读取十字测温点的温度值，形成第二温度序列；
步骤S107：以第一温度序列为输入，以第二温度序列为输出，训练虚拟十字测温点和真实十字测温点的温度关系模型。
进一步的，在步骤S10中，所述高炉的平面坐标系，设定高炉炉顶坐标原点位于十字测温装置测温点所在水平面的高炉中心。
进一步的，所述步骤S103包括：
选取高炉内炉壳的圆周上的四个点的坐标为A(-R,0),B(R.0),C(0,R),D(0,-R)，其中R是高炉的内径，这四个点在红外图像上的坐标通过手动测量或者图像算法计算上下左右的边界点得到A’(XA,YA),B’(XB,YB),C’(XC,YC),D’(XD,YD)；
根据透视变换原理，计算出从高炉坐标系到红外图像I1坐标系的透视变换矩阵P1，并根据透视变换矩阵得到原始红外图像I1转换到高炉的水平坐标系的鸟瞰温度图像I2。
进一步的，所述步骤S107中，所述温度关系模型基于MLP神经网络模型。
进一步的，所述步骤S20包括：
步骤S201：在高炉炉顶安装红外测温装置，获取炉顶的红外图像；
步骤S202：根据红外测温装置在炉顶的安装位置，建立透视变换矩阵，将红外图像I3校正为位于高炉平面坐标系的鸟瞰温度图像I4；
步骤S203：设定十字测温点的位置；
步骤S204：从鸟瞰温度图像I4中找到与十字测温点对应的像素点坐标，再通过透视变换矩阵，在红外图像I3建立虚拟十字测温点；
步骤S205：提取虚拟十字测温点的温度值，形成第三温度序列；
步骤S206：将第三温度序列输入已训练的虚拟十字测温点和真实十字测温点的温度关系模型，输出各十字测温点的预测温度值组成的第四温度序列。
本发明实现了如下技术效果：
采用本方法，无需安装传统的十字测温装置，通过红外测温装置获取原始红外图像，将原始红外图像的测温结果转换为十字测温点的温度值，并利用现 有的十字测温温度模型执行相应的高炉操作和决策。
附图说明
图1是本发明的基于红外测温的高炉软十字测温方法的流程图；
图2是本发明的红外图像和鸟瞰温度图像的位置映射图；
图3是本发明涉及的神经网络模型。
具体实施方式
为进一步说明各实施例，本发明提供有附图。这些附图为本发明揭露内容的一部分，其主要用以说明实施例，并可配合说明书的相关描述来解释实施例的运作原理。配合参考这些内容，本领域普通技术人员应能理解其他可能的实施方式以及本发明的优点。
现结合附图和具体实施方式对本发明进一步说明。
如图1所示，本申请提出了一种基于红外测温的高炉软十字测温方法，包括以下步骤：
步骤S10，在高炉上同时设置红外测温装置和十字测温装置；根据红外图像的像素点和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点；并根据虚拟十字测温点的温度值和十字测温点的温度值，训练虚拟十字测温点和真实十字测温点的温度关系模型。
步骤S20，在高炉上仅设置红外测温装置；设定十字测温点的位置，根据红外图像的像素点和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点的位置；从红外图像提取虚拟十字测温点的温度值，再通过温度关系模型，输出十字测温点的预测温度值。
具体的，步骤S10包括：
在高炉上同时安装十字测温装置和红外测温装置；根据十字测温装置在炉内的安装平面，建立高炉的平面坐标系；根据红外测温装置在炉顶的安装位置，建立透视变换矩阵，将红外图像I1校正为位于高炉平面坐标系的鸟瞰温度图像 I2；根据十字测温装置在炉内的安装位置，在鸟瞰温度图像I2中找到对应的像素点坐标，根据这些像素点坐标建立虚拟十字测温点。读取虚拟十字测温点的温度值，形成第一温度序列；通过十字测温装置读取十字测温点的温度值，形成第二温度序列。
具体的，在本实施例中，给出了如上步骤的具体实现方案：由于红外测温相机并不能安装在高炉炉顶的中心线上，完全平行于水平面拍照，而是安装在高炉侧壁上与水平面呈一定的夹角，因此十字测温的同心圆图案在红外图像上并不是圆形，而是同心圆经过透视变换后的图案，如图2所示。因此第一步需求解从高炉实际坐标到红外图像坐标上的透视变换矩阵P。
建立高炉坐标系，设高炉炉顶坐标原点位于十字测温装置安装平面(即十字测温装置的测温点所在的水平面)的高炉中心上，如图2所示高炉内炉壳的圆周上的四个点的坐标为A(-R,0),B(R.0),C(0,R),D(0,-R)，其中R是高炉的内径，这四个点在红外图像上的坐标可以通过手动测量或者图像算法计算上下左右的边界点得到A’(XA,YA),B’(XB,YB),C’(XC,YC),D’(XD,YD)，这里以图像左上角的点为原点。根据透视变换原理，可以计算出从高炉坐标系到红外图像I1坐标系的透视变换矩阵P1，并根据透视变换矩阵得到原始红外图像I1转换到高炉的水平坐标系的鸟瞰温度图像I2。
由此可以设计出一个虚拟的十字测温方案，比如在0°,90°，180°，270°这4个方向上每间隔0.8米设置一个测温点，每个方向设置5个点。测温点的坐标为：
Figure PCTCN2022122616-appb-000001
则它在红外图像I3上的对应点为
Figure PCTCN2022122616-appb-000002
Figure PCTCN2022122616-appb-000003
为虚拟十字测温点的位置坐标。
该虚拟的十字测温方案可采用原有的十字测温方案，从而便于利用原有的温 度模型进行高炉操作和决策。也可以设定新的十字测温方案，并训练新的温度模型以支持高炉操作和决策。
然后计算二者温度的转换关系，考虑到两者的转换关系可能是非线性，可以设计了一个包含多个隐藏层的神经网络模型来表征这一关系，网络的输入层为红外测温的结果，输出层为拟合的对应的十字测温点的测温结果，网络包含2个隐藏层，每个隐藏层包含10个节点，如图3所示。
在本实施例中，给出了如下虚拟十字测温点和真实十字测温点的温度关系模型训练方案：以2小时为间隔分别采样20个十字测温点和其对应的红外测温的结果，它们组成一个样本数据，本实施方案采样了50天共12000组样本数据对温度修正网络进行交叉验证求解。我们随机将80％的数据作为训练集用来训练温度校正网络，10％的数据作为验证集来控制训练周期数，10％的数据作为测试集来测试模型的泛化效果。损失函数采用平均误差平方和(Mean Squared Error)，其定义为：
Figure PCTCN2022122616-appb-000004
其中
M为某个样本集合的样本数，样本集合包括训练集、验证集和测试集；
T(k),T S(k)分别为第k组测量的真实十字测温值和网络输出的测温值。
采用Adam优化算法训练100个周期后验证集误差变化趋于0后终止训练，将权重和偏置参数保存至模型文件Model。
在具体实施过程中，为实现虚拟十字测温点和真实十字测温点的温度关系模型训练，可以采用MLP(多层感知机)等神经网络模型进行训练。MLP多层感知机是一种前向结构的人工神经网络ANN，映射一组输入向量到一组输出向量。MLP可以被看作是一个有向图，由多个节点层组成，每一层全连接到下一层。除了输入节点，每个节点都是一个带有非线性激活函数的神经元。
具体的，步骤20包括：
步骤S201：在高炉炉顶安装红外测温装置；
步骤S202：根据红外测温装置在炉顶的安装位置，建立透视变换矩阵，将红外图像I3校正为位于高炉平面坐标系的鸟瞰温度图像I4；
步骤S203：设定十字测温点的位置；
步骤S204：从鸟瞰温度图像I4中找到与十字测温点对应的像素点坐标，再通过透视变换矩阵，在红外图像I3建立虚拟十字测温点；
步骤S205：提取虚拟十字测温点的温度值，形成第三温度序列；
步骤S206：将第三温度序列输入已训练的虚拟十字测温点和真实十字测温点的温度关系模型，输出各十字测温点的预测温度值组成的第四温度序列。
第四温度序列为十字测温点的预测温度值，根据这些预测温度值，可利用现有的十字测温的温度模型执行相应的高炉操作和决策。
尽管结合优选实施方案具体展示和介绍了本发明，但所属领域的技术人员应该明白，在不脱离所附权利要求书所限定的本发明的精神和范围内，在形式上和细节上可以对本发明做出各种变化，均为本发明的保护范围。

Claims (6)

  1. 一种基于红外测温的高炉软十字测温方法，其特征在于，包括以下步骤：
    步骤S10，在高炉上同时设置红外测温装置和十字测温装置；根据红外图像的像素点的位置和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点；并根据虚拟十字测温点的温度值和十字测温点的温度值，训练虚拟十字测温点和真实十字测温点的温度关系模型；
    步骤S20，在高炉上仅设置红外测温装置；设定十字测温点的位置，根据红外图像的像素点和十字测温点的位置映射关系，确定红外图像上的虚拟十字测温点；从红外图像提取虚拟十字测温点的温度值，再通过温度关系模型，输出十字测温点的预测温度值。
  2. 如权利要求1所述的基于红外测温的高炉软十字测温方法，其特征在于，所述步骤S10，包括：
    步骤S101：在高炉上同时安装十字测温装置和红外测温装置；
    步骤S102：根据十字测温装置在炉内的安装平面，建立高炉的平面坐标系；
    步骤S103：根据红外测温装置在炉顶的安装位置，建立透视变换矩阵，将红外图像I1校正为位于高炉平面坐标系的鸟瞰温度图像I2；
    步骤S104：根据十字测温装置在炉内的安装位置，在鸟瞰温度图像I2中找到十字测温点对应的像素点坐标，通过这些像素点坐标，在红外图像I1找到对应的像素点坐标，建立虚拟十字测温点；
    步骤S105：读取虚拟十字测温点的温度值，形成第一温度序列；
    步骤S106：通过十字测温装置读取十字测温点的温度值，形成第二温度序列；
    步骤S107：以第一温度序列为输入，以第二温度序列为输出，训练虚拟十字测温点和真实十字测温点的温度关系模型。
  3. 如权利要求2所述的基于红外测温的高炉软十字测温方法，其特征在于，在步骤S10中，所述高炉平面坐标系，设定高炉炉顶坐标原点位于十字测温安 装高度的平面的高炉中心。
  4. 如权利要求3所述的基于红外测温的高炉软十字测温方法，其特征在于，所述步骤S103包括：
    选取高炉内炉壳的圆周上的四个点的坐标为A(-R,0),B(R.0),C(0,R),D(0,-R)，其中R是高炉的内径，这四个点在红外图像上的坐标通过手动测量或者图像算法计算上下左右的边界点得到A’(XA,YA),B’(XB,YB),C’(XC,YC),D’(XD,YD)；
    根据透视变换原理，计算出从高炉坐标系到红外图像I1坐标系的透视变换矩阵P1，并根据透视变换矩阵得到原始红外图像I1转换到高炉的水平坐标系的鸟瞰温度图像I2。
  5. 如权利要求2所述的基于红外测温的高炉软十字测温方法，其特征在于，所述步骤S107中，所述温度关系模型基于MLP神经网络模型。
  6. 如权利要求1所述的基于红外测温的高炉软十字测温方法，其特征在于，所述步骤S20，包括：
    步骤S201：在高炉炉顶安装红外测温装置；
    步骤S202：根据红外测温装置在炉顶的安装位置，建立透视变换矩阵，将红外图像I3校正为位于高炉平面坐标系的鸟瞰温度图像I4；
    步骤S203：设定十字测温点的位置；
    步骤S204：从鸟瞰温度图像I4中找到与十字测温点对应的像素点坐标，再通过透视变换矩阵，在红外图像I3建立虚拟十字测温点；
    步骤S205：提取虚拟十字测温点的温度值，形成第三温度序列；
    步骤S206：将第三温度序列输入已训练的虚拟十字测温点和真实十字测温点的温度关系模型，输出各十字测温点的预测温度值组成的第四温度序列。
PCT/CN2022/122616 2022-09-21 2022-09-29 一种基于红外测温的高炉软十字测温方法 WO2024060285A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202211165053.XA CN115597715A (zh) 2022-09-21 2022-09-21 一种基于红外测温的高炉软十字测温方法
CN202211165053.X 2022-09-21

Publications (1)

Publication Number Publication Date
WO2024060285A1 true WO2024060285A1 (zh) 2024-03-28

Family

ID=84844423

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/122616 WO2024060285A1 (zh) 2022-09-21 2022-09-29 一种基于红外测温的高炉软十字测温方法

Country Status (2)

Country Link
CN (1) CN115597715A (zh)
WO (1) WO2024060285A1 (zh)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102732660A (zh) * 2012-06-27 2012-10-17 浙江大学 一种基于多源信息融合的料面温度场检测方法
CN105002321A (zh) * 2015-06-16 2015-10-28 内蒙古科技大学 一种处理煤气流中心动态跟踪及煤气利用率监测方法
CN106227699A (zh) * 2016-07-04 2016-12-14 东北大学 一种高炉炉喉十字测温装置中心带温度预测方法及系统
US20190093186A1 (en) * 2017-09-27 2019-03-28 International Business Machines Corporation Manufacturing process control with deep learning-based predictive model for hot metal temperature of blast furnace

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102732660A (zh) * 2012-06-27 2012-10-17 浙江大学 一种基于多源信息融合的料面温度场检测方法
CN105002321A (zh) * 2015-06-16 2015-10-28 内蒙古科技大学 一种处理煤气流中心动态跟踪及煤气利用率监测方法
CN106227699A (zh) * 2016-07-04 2016-12-14 东北大学 一种高炉炉喉十字测温装置中心带温度预测方法及系统
US20190093186A1 (en) * 2017-09-27 2019-03-28 International Business Machines Corporation Manufacturing process control with deep learning-based predictive model for hot metal temperature of blast furnace

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
XU YONG-HUA, WU MIN, CAO WEI-HUA, NING ZHI-YU: "Measurement of the Temperature Profile Based on Infrared Image Processing and its Application in Blast Furnace", CONTROL ENGINEERING OF CHINA, CN, vol. 12, no. 4, 31 July 2005 (2005-07-31), CN, pages 354 - 356, XP009553403, ISSN: 1671-7848 *

Also Published As

Publication number Publication date
CN115597715A (zh) 2023-01-13

Similar Documents

Publication Publication Date Title
US20200355601A1 (en) Thermography image processing with neural networks to identify corrosion under insulation (cui)
Lagüela et al. Automation of thermographic 3D modelling through image fusion and image matching techniques
CN111337142A (zh) 体温修正方法、装置及电子设备
US20210018426A1 (en) Metal loss determinations based on thermography machine learning approach for insulated structures
Protopapadakis et al. Autonomous robotic inspection in tunnels
CN107843818B (zh) 基于异源图像温升与局部放电特征高压绝缘故障诊断方法
WO2019051961A1 (zh) 一种管道检测方法、装置以及存储介质
CN112102271B (zh) 一种基于普通数字相机的实时在线火焰温度测量方法
CN108917205B (zh) 用于太阳能接收器的热-机械监测的系统和方法
CN111103296B (zh) 一种建筑外墙面层质量的非接触式检测系统的数据测量控制处理装置
Xiong et al. Passive vision measurement for robust reconstruction of molten pool in wire and arc additive manufacturing
US20150355030A1 (en) Equipment and method for intensity-temperature transformation of imaging system
EP4341223A1 (en) Method of control, control system and glass furnace, in particular for temperature/thermal control
CN105828034B (zh) 一种管式反应炉炉膛全景图像成像方法
Dabetwar et al. Sensitivity analysis of unmanned aerial vehicle-borne 3D point cloud reconstruction from infrared images
CN114659523B (zh) 一种大范围高精度姿态测量方法及装置
CN114119614A (zh) 一种远程检测建筑物的裂缝的方法
WO2024060285A1 (zh) 一种基于红外测温的高炉软十字测温方法
KR102220654B1 (ko) 측정온도 보정시스템
CN113077447A (zh) 一种可探测高压设备微弱放电并进行缺陷分析的方法
CN117473862A (zh) 高炉料面温度场检测方法
Jo et al. Dense thermal 3d point cloud generation of building envelope by drone-based photogrammetry
CN115797463A (zh) 一种fod雷达与相机的神经网络训练方法及全局标定方法
CN211453394U (zh) 一种建筑外墙面层质量的非接触式检测系统
CN114494245B (zh) 墙体结构孔气密性检测的图像识别方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 22959287

Country of ref document: EP

Kind code of ref document: A1