WO2020010937A1

WO2020010937A1 - Machine vision-based model predictive control algorithm for material layer thickness of grate cooler

Info

Publication number: WO2020010937A1
Application number: PCT/CN2019/088672
Authority: WO
Inventors: 张成伟; 刘小蒙; 李慧霞; 任静
Original assignee: 南京凯盛国际工程有限公司
Priority date: 2018-07-11
Filing date: 2019-05-27
Publication date: 2020-01-16
Also published as: CN108800916A; CN108800916B

Abstract

Provided is a machine vision-based model predictive control method for material layer thickness of a grate cooler, comprising the following steps: Step one, software and hardware deployment; Step two, measuring the thickness of the material layer by using machine vision technology; Step three, offline recognition; and Step four, online control. The thickness of the material layer is directly measured through the machine vision, which avoids the problems such as inaccuracy and large interference caused by indirect characterization measurement. The model predictive control algorithm can stably control the material layer thickness, maintain the uniform distribution of the material layer, improve the heat exchange efficiency, and save energy.

Description

A Machine Vision Based Prediction Control Algorithm for the Layer Thickness Model of Grate Cooler

Technical field

The invention relates to a predictive control algorithm for a grate cooler material layer thickness model based on machine vision.

Background technique

Grate cooler is the key equipment in cement production, mainly composed of grate plate and grate bed. The high-temperature clinker is transported from the rotary kiln to the grate cooler grate bed. The grate plate pushes the clinker out, and moves forward along the grate bed to push the clinker. After that, the grate plate returns to transport the clinker again, and so on. The cooling air under the grate bed passes through the clinker to cool it. Part of the cooling air is heated to form secondary air and enters the rotary kiln, and part of the air enters the waste heat power generation device.

The thickness of the grate cooler material layer is very important for the grate cooler equipment. If the material layer is too thin, the time for the cooling air to pass through the material layer is too short and a short circuit is formed. The cooling air passing through the material layer has fast air speed and large air volume. The temperature of the secondary air is reduced; the material layer is too thick, and the cooling air is impermeable, which increases the downforce pressure, and then there is a partial blow-through. The cooling air is blown away from the blow-through position, and the "red river" phenomenon is likely to occur in other positions, resulting in The grate plate is locally overheated, and the thickness of the material layer is increased, so that the downforce pressure is increased. The high-pressure cold air used for heat exchange is blocked, and the amount of air entering the grate chamber is reduced, resulting in lower secondary air temperature and reduced heat transfer efficiency. Therefore, the control of the grate cooler is most important to make the thickness of the material layer as thick as possible and the whole of the grate bed to be in a boiling state. The thickness of the material layer is adjusted by the speed of the slab motion (hereinafter referred to as the slag speed). The slab speed is faster. The clinker quickly passes through the slab bed, and the thickness of the slab layer will decrease. If the slab speed is slow, the clinker will accumulate on the slab bed. The thickness will increase, and the thickness of the material layer will show a certain integral characteristic and will always increase. In addition, the thickness of the material layer is also affected by the cooling air volume. If the air volume is large, the clinker is cooled faster and the clinker viscosity is reduced, so that the material layer is easily pushed out of the grate cooler. The thickness of the material layer is reduced. Being cooled slowly, the clinker has a high viscosity, which makes it difficult for the clinker to be pushed out, and the thickness of the clinker increases.

Since the grate cooler has a harsh internal environment and is relatively closed, it is difficult to directly measure the thickness of the material layer inside. How to determine the appropriate material layer thickness has always been a difficult point for grate cooler optimization. At present, the thickness of the material layer is generally characterized by indirect quantities related to the thickness of the material layer (downward pressure, secondary air temperature, hydraulic signal of the grate cooler, etc.). The traditional control strategy considers that the indirect quantity and the thickness of the material layer meet a linear relationship. A model between the indirect quantity and the velocity is established, and the control is implemented through model predictive control or fuzzy control. Because the model of indirect quantity and material layer thickness is more complicated, it shows non-linear characteristics, and the relationship between indirect quantity and material layer thickness is greatly affected by the working conditions, which causes great troubles for the optimization control of material layer thickness, resulting in the current market. The optimization of the cold machine cannot achieve the benefit improvement, and human intervention is required when the working conditions change.

Summary of the invention

In order to overcome the shortcomings of the prior art, the present invention provides a predictive control algorithm for the thickness model of the grate cooler based on machine vision. The algorithm uses machine vision technology to measure the thickness of the material in real time, and establishes the relationship between the thickness of the material and the speed of the grate. The model relationship, meanwhile, takes the secondary air temperature and the downhill pressure as constraint variables, and uses the model predictive control algorithm to solve the slowdown speed to form a closed-loop control loop. This algorithm can effectively solve the problem of inaccurate indirect characterization of the material layer thickness of the grate cooler, and provides accurate data information for the material layer thickness. Finally, it can achieve stable control of the material layer thickness, improve the quality of clinker, improve heat exchange efficiency, and save energy. Consumption purpose.

The technical solution adopted by the present invention for solving the above technical problems is: a predictive control algorithm for the thickness model of the grate cooler material layer based on machine vision, including the following steps;

(1) Install an industrial camera above the exit of the grate cooler material layer; install a ruler at the side of the grate cooler material layer exit, deploy machine vision software at the engineer station, and receive it through industrial communication protocols (TCP / IP, OPC, etc.) Two-dimensional image data transmitted by industrial cameras; install expert optimization system software independently developed by Nanjing Kaisheng;

(2) Define the ROI area, convert the RGB image into a gray value image, extract the edge contour of the clinker by smoothing, filtering, and edge detection, and calculate the thickness of the material layer by scaling conversion with the size of the ruler;

(3) The calculated thickness of the material layer is sent to the OPC server through OPC to record the historical trend of the material layer thickness; at the same time, the historical trend of the speed, secondary air temperature, and downforce pressure is recorded; the speed is considered as a manipulated variable (MV), the thickness of the material layer is used as the controlled variable (CV), the secondary air temperature and the downforce pressure are used as the constraint variables (CCV), an integration model of one input and three outputs is established, and an expert optimization system is used for system identification;

(4) Use the model identified in step (3) to implement model predictive control (MPC) in the expert optimization system. The MPC algorithm performs control through model prediction, rolling optimization, and feedback correction cycles;

Preferably, the objective function is subjected to rolling optimization during the control period in each of the steps (4).

Preferably, a feedback correction is performed on the predicted value calculated by the rolling optimization.

Compared with the prior art, the advantages of the present invention are:

First, the present invention measures the thickness of the grate cooler through a single-camera and ruler comparison method using a machine vision algorithm, which avoids the complicated and costly processes such as three-dimensional modeling and measurement of binocular cameras commonly used in the market, and infrared measurement. High measurement method.

Secondly, the present invention directly measures the thickness of the material layer, which avoids the influence of uncertain factors such as inaccurate models and environmental disturbances when traditional indirect quantities are used to make the measurement results more accurate.

Third, the present invention performs model predictive control by directly measuring the thickness of the material layer, which can keep the material layer thickness as thick as possible without local eruption, and the material layer is maintained in an optimal boiling state, so that the secondary air temperature is mild. Downforce pressure meets process requirements, improves heat exchange efficiency, and achieves a win-win goal of quality and environmental protection.

Finally, the present invention does not need to modify field equipment except for industrial cameras, so it is easy to deploy. This algorithm opens a new way for grate cooler material layer thickness measurement and optimization control.

BRIEF DESCRIPTION OF THE DRAWINGS

1 is a schematic diagram of an industrial camera installation according to the present invention;

FIG. 2 is an interface diagram of Nanjing Kaisheng expert optimization system in the present invention; FIG.

3a is an original image of an image processing process diagram of the present invention;

3b is an ROI region image of an image processing process diagram of the present invention;

FIG. 3c is an image after graying processing of the image processing process diagram of the present invention; FIG.

FIG. 3d is an image after the contour is extracted from the image processing process diagram of the present invention;

FIG. 4 is a communication and control flowchart of the present invention.

detailed description

A method for predicting and controlling a layer thickness model of a grate cooler based on machine vision, including the following steps;

Step one: software and hardware deployment; install an industrial camera above the grate cooler material layer exit side, protect the camera from high temperature interference by a protective cover; install a ruler on the side of the grate cooler material layer exit to calibrate the thickness of the material layer, The installation diagram of the industrial camera and ruler is shown in Figure 1. In order to capture an accurate material layer image of the grate cooler, the industrial camera is installed above the grate cooler exit side, so that the grate cooler can be viewed from a larger angle. The internal state, in addition to installing an industrial protective cover to prevent the industrial camera from being affected by high temperature and dust; install a scale at the outlet of the grate cooler, as shown by the blue vertical bar in Figure 1. It is generally believed that the thickness of the material layer is equal to the scale. Linear relationship, the actual material layer thickness can be obtained through the image information of the material layer collected and converted by the scale after image processing; machine vision software is deployed at the engineer station, and industrial camera protocols (TCP / IP, OPC, etc.) are used to receive the industrial camera transmission 2D image data; install the expert optimization system software independently developed by Nanjing Kaisheng, which integrates industrial applications such as identification, model predictive control algorithms, etc. The control algorithm used, the interface of the expert optimization system is shown in Figure 2. Nanjing Kaisheng expert optimization system covers model predictive control, fuzzy control, generalized predictive control and other control algorithms. It has scripting technologies such as smoothing and filtering, and can browse the control in real time. Curves (such as the manipulated variable MV and the controlled variable CV trend in the figure) are convenient for adjusting control parameters and setting variable values. When the model predictive control algorithm is actually used, the manipulated variable MV, the controlled variable CV, etc. are configured to the relevant OPC variables, and the parameter configuration function can be used to achieve the system optimal control.

Step 2: Apply machine vision technology to measure the thickness of the clinker. Because the temperature in the upper part of the grate cooler is high, the upper part of the clinker is in a molten state, so it is red in the image, and the unfired clinker at the bottom is dark black; Therefore, the ROI area can be delineated to convert the RGB image into a grayscale image, and the edge contour of the clinker can be extracted by smoothing, filtering, and edge detection. The image processing process is shown in Figure 3 (a ~ d); It is distributed in a medium proportion with the ruler in the camera, so the thickness of the material layer can be calculated by proportional conversion with the size of the ruler; the image processing of the material layer thickness of the grate cooler mainly includes four parts. The first is the original image acquisition part. The original graphics are collected by the industrial camera and transmitted to the machine vision software installed on the engineer station through the industrial communication protocol. The second is the machine vision software to analyze and process the obtained two-dimensional image information to obtain the ROI area. The third is the graying process of the ROI region; the fourth is to extract the contour of the material layer from the gray image. After obtaining the outline of the material layer, the actual material layer thickness can be obtained by converting with the scale.

Step 3: Offline identification; send the calculated material layer thickness to the OPC server via OPC, and record the historical trend of material layer thickness; record the historical trend of sag speed, secondary air temperature, and sag pressure at the same time; The manipulated variable (MV) is used, the thickness of the material layer is used as the controlled variable (CV), the secondary air temperature and the downforce pressure are used as the constraint variables (CCV). An integration model of one input and three outputs is established, and the system is identified by an expert optimization system.

Step 4. Online control. Use the model identified in Step 3 to implement model predictive control (MPC) in an expert optimization system. The MPC algorithm performs control through model prediction, rolling optimization, and feedback correction cycles.

In each control cycle, the objective function is rolled and optimized; the purpose of the optimized performance index is: to ensure that the thickness of the material layer is as close as possible to the set value, and the secondary air temperature and the downforce pressure are within the upper and lower constraints. Avoid rapid changes in speed.

The predicted value calculated by rolling optimization is corrected by feedback; due to unknown factors such as model mismatch and environmental interference in actual operation, the predicted value calculated by rolling optimization may deviate from the actual value. If real-time information is not used for timely correction of feedback The next optimization will be based on inaccurate model predictions. As the process progresses, the predicted output may deviate more and more from the actual output.

The communication part of the invention mainly relates to the communication between the industrial camera and the machine vision software, the machine vision software and the OPC server, the OPC client and the OPC server. Industrial cameras and machine vision software follow standard industrial communication protocols, such as TCP / IP, OPC, DA, OPC, UA, etc .; machine vision software and OPC Client and OPC Server use OPC protocol for communication. The first part of the control process is to establish a prediction model between the manipulated variable and the controlled quantity. This process can be obtained by offline model identification. Secondly, the controlled system is optimized in each control cycle to ensure the thickness of the material layer. Keep it as thick and evenly distributed as possible, and avoid frequent and severe vibrations at the fast speed. Finally, the feedback correction of the predicted output value is made up to compensate for the problem of control accuracy degradation caused by environmental interference and distortion of the prediction model.

Since the grate cooler has a harsh internal environment and is relatively closed, it is difficult to directly measure the thickness of the material layer inside. How to determine the appropriate material layer thickness has always been a difficult point for grate cooler optimization. At present, the thickness of the material layer is generally characterized by indirect quantities related to the thickness of the material layer (downward pressure, secondary air temperature, hydraulic signal of the grate cooler, etc.). The traditional control strategy considers that the indirect quantity and the thickness of the material layer meet a linear relationship, establish a model between the indirect quantity and the speed, and implement control through model predictive control or fuzzy control. Because the model of indirect quantity and material layer thickness is more complicated, it shows non-linear characteristics, and the relationship between indirect quantity and material layer thickness is greatly affected by the working conditions, which causes great troubles for the optimization control of material layer thickness, resulting in the current market. The optimization of the cold machine cannot achieve the benefit improvement, and human intervention is required when the working conditions change.

The invention discloses a predictive control algorithm for a grate cooler material layer thickness model based on machine vision. The algorithm uses an industrial camera to take a two-dimensional image of the inside of the grate cooler and transfers it to the machine vision software. The machine vision software obtains the ROI region from the two-dimensional image, extracts the edge contour of the material layer through graying, smoothing, filtering, and edge detection, and converts it with the scale in the image to obtain the actual thickness of the material layer. The thickness of the material layer is used as the controlled quantity, the secondary air temperature and the down pressure are used as the constraint variables, and the speed is used as the manipulated variable. The system model is obtained through offline calculation. The system model established offline is introduced into the expert optimization system developed by Nanjing Kaisheng to implement online model predictive control. The thickness of the material layer is directly measured by machine vision, which avoids problems such as inaccuracy and large interference caused by indirect quantity characterization. The model predictive control algorithm can stably control the thickness of the material layer, keep the material layer uniformly distributed, improve heat exchange efficiency, and save energy. The algorithm is easy to deploy and simple to maintain, which opens a new way for the measurement and control of the thickness of the grate cooler material layer.

In the end, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, but not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still The technical solutions described in the foregoing embodiments are modified, or some technical features are replaced equivalently; and these modifications or replacements do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

A machine vision-based predictive control algorithm for the thickness layer model of a grate cooler, including the following steps;

(1) Install an industrial camera above the exit of the grate cooler material layer; install a ruler at the side of the grate cooler material layer exit, deploy machine vision software at the engineer station, and receive the two-dimensional image data transmitted by the industrial camera through the industrial communication protocol ; Install expert optimization system software;

(2) Define the ROI area, convert the RGB image into a gray value image, extract the edge contour of the clinker by smoothing, filtering, and edge detection, and calculate the thickness of the material layer by scaling conversion with the size of the ruler;

(3) The calculated thickness of the material layer is sent to the OPC server through OPC to record the historical trend of the material layer thickness; at the same time, the historical trend of the speed, secondary air temperature, and downforce pressure is recorded; the speed is considered as a manipulated variable (MV), the thickness of the material layer is used as the controlled variable (CV), the secondary air temperature and the downforce pressure are used as the constraint variables (CCV), an integration model of one input and three outputs is established, and an expert optimization system is used for system identification;

(4) Use the model identified in step (3) to implement model predictive control (MPC) in the expert optimization system. The MPC algorithm performs control through model prediction, rolling optimization, and feedback correction cycles;
The predictive control algorithm for the thickness layer model of the grate cooler based on machine vision according to claim 1, characterized in that, in each control step in the step (4), the objective function is subjected to rolling optimization.
The predictive control algorithm for the thickness layer model of the grate cooler based on machine vision according to claim 2, characterized in that a feedback correction is performed on the predicted value calculated by the rolling optimization.