An intelligent material flow detection method and system based on ranging radar technology

Abstract

An intelligent material flow detection method and system based on ranging radar technology, the method comprising: collecting material flow state data, constructing a belt material flow state original data set; obtaining a belt material flow state data set by pre-processing and filtering junk data, and then dividing it into a training set, a validation set and a test set according to a certain ratio, and labeling the training set and the validation set; using the belt material flow state data set to train a material flow state classification model through the ML.NET framework; calculating the macro accuracy of the material flow state classification model, and using the macro accuracy as a performance indicator for evaluation to obtain an optimal material flow state classification model; processing real-time material flow state data through the optimal material flow state classification model to obtain material flow state classification, using a preset sliding window to receive normal material flow state data, and updating the belt material flow state according to the consistency of the data state in the sliding window; the present invention reduces the overall cost while achieving high-precision detection, and improves system reliability and long-term operation stability.

Description

Technical Field

[0001] The present invention relates to the technical field of material flow detection, and in particular to an intelligent material flow detection method and system based on ranging radar technology. Background Art

[0002] The existing technical solutions for belt conveyor material flow detection are mainly contact material flow sensing technology solutions and non-contact material flow sensing technology solutions:

[0003] The contact material flow sensing technical solution takes the utility model patent with publication number CN217331212U and the name of a material flow detection device as an example. The patent solution includes: two symmetrically arranged brackets (1) are arranged above the belt conveyor, and the lower surface thereof is fixedly connected to the upper surface of the belt conveyor. The adjacent side walls of the two brackets (1) are rotatably connected to the same rotating shaft (2), and the peripheral side of the rotating shaft (2) is fixedly connected to a first detection rod (3). The lower half side wall of the first detection rod (3) is fixedly connected to a second detection rod (4), and the upper surface of the first detection rod (3) is fixedly connected to a trigger plate (5). The two brackets (1) are fixedly connected to the same support plate (6), and the upper surface of the support plate (6) is fixedly connected to an element bracket (7), and the side wall of the element bracket (7) is provided with a detection element (8);

[0004] When the belt conveyor starts to transport materials, the materials first come into contact with the second detection rod 4. When the materials come into contact with the second detection rod 4, the first detection rod 3 will be driven to rotate. When the first detection rod 3 is rotated, the rotating shaft 2 will rotate at the same time. At the same time, when the first detection rod 3 rotates, the trigger plate 5 will be driven to rotate. When the trigger plate 5 rotates, it will pass through the detection element 8. At this time, the detection element 8 will be triggered. The detection element 8 is electrically connected to the external detection device. By setting the detection element 8, it is convenient to trigger the detection device and facilitate the user to obtain messages.

[0005] The non-contact material flow detection technical solution takes the invention application with publication number CN118597712A and the name of a belt conveyor material flow detection device and its detection method as an example. The detection device of the invention application solution includes: a fixed frame 1, a horizontal adjustment frame 2 and a vertical adjustment frame 3. The fixed frame 1 is fixed on the bracket 5 on both sides of the belt 4. The horizontal adjustment frame 2 is connected to the top of the fixed frame 1 through a positioning and adjustment device. Multiple vertical adjustment frames 3 are respectively arranged at different positions of the horizontal adjustment frame 2. Any vertical adjustment frame 3 is provided with a sensor for detecting material 6, and any sensor is arranged directly above the material 6. The controller is connected to the multiple sensors respectively, and in response to the signals fed back by the multiple sensors that the material 6 passes, the controller starts the water pump to perform watering operations.

[0006] The detection method includes: in response to the start signal of the belt conveyor, the controller controls the multiple sensors to start working; in response to the signals fed back by all the sensors indicating that materials have passed, the controller starts the water pump to perform watering operations; if the controller does not receive any signal fed back by any of the sensors indicating that materials have passed, an early warning is issued through the alarm system.

[0007] For the contact material flow detection solution, since this solution must in principle cause the detection device to have direct physical contact with the material flow transported at a certain speed on the belt, long-term operation will inevitably lead to wear of the mechanical structure, resulting in a gap in detection accuracy and stability compared to the non-contact material flow detection solution, and will also increase the user's maintenance workload in daily maintenance.

[0008] For non-contact material flow detection solutions, the current selection of sensors is mainly based on laser radar or cameras. As we all know, the material flow detection system based on laser radar generally uses linear laser radar, and its main advantage is that it can calculate the material flow rate based on the detection results and the belt running speed; however, its advantages are not consistent with the application scenarios that only require material flow status detection, which results in high cost waste; and the computer vision detection and recognition solution using cameras is not only expensive, but also has poor detection effect when facing coal flow, which is visually similar to the material flow when the belt is empty, and is easily affected by environmental factors such as light, and faces the problem of target detection model optimization.

[0009] Therefore, it is necessary to invent a low-cost, high-precision and high-stability detection method and system that only detects the flow state. Summary of the invention

[0010] In view of the above problems, the present invention proposes an intelligent material flow detection method and system based on ranging radar technology, and the technical solutions adopted are as follows:

[0011] An intelligent material flow detection method based on ranging radar technology, characterized in that it comprises the following steps:

[0012] The material flow state data of the belt conveyor when it is working is collected in units of a single set of distance and signal strength, and the original data set of the belt material flow state is constructed;

[0013] Preprocess the original data set of belt material flow status, filter out junk data, obtain the belt material flow status data set, divide the belt material flow status data set into training set, verification set and test set, and annotate the training set and verification set data;

[0014] The belt material flow status dataset is used to train the material flow status classification model through the ML.NET framework;

[0015] Calculate the macro accuracy of the material flow state classification model, and use the macro accuracy as a performance indicator to evaluate the material flow state classification model to obtain the optimal material flow state classification model;

[0016] The material flow state classification is obtained by processing the real-time material flow state data through the optimal material flow state classification model, a preset sliding window is used to receive normal material flow state data, and the belt material flow state is updated according to the consistency of the data state in the sliding window.

[0017] Furthermore, the preprocessing of the original data set of the belt material flow state includes: analyzing the mean, variance, and peak value based on the distance and signal strength data, filtering out junk data that affects the training of the material flow state classification model, and improving the quality of the belt material flow state data set.

[0018] Furthermore, before the belt material flow state data set is used to train the material flow state classification model through the ML.NET framework, the method includes:

[0019] According to the distance and signal strength characteristics of the belt under different material flow states, the data of the training set and the verification set in the belt material flow state data set are divided into three states: with material, without material and abnormal data.

[0020] Furthermore, the belt material flow state data set is used to train the material flow state classification model through the ML.NET framework, including:

[0021] The ML.NET framework is used to train the belt material flow state dataset using the randomized dual coordinate ascent (SDCA) algorithm, limited memory BFGS (L-BFGS) algorithm, lightweight gradient boosting machine (LightGBM) algorithm, fast decision tree algorithm and fast decision forest algorithm.

[0022] Furthermore, the macro accuracy of the material flow state classification model is calculated, and the macro accuracy is used as a performance indicator to evaluate the material flow state classification model to obtain the best material flow state classification model, including:

[0023] There are three types of belt material flow states: material, no material, and abnormal. Let A1, A2, and A3 be the accuracy rates of material, no material, and abnormal categories respectively, then the macro accuracy is expressed as:

[0024]

[0025] The closer the macro accuracy is to 1, the better the classification performance of the material flow state classification model for all categories. It is used as an evaluation optimization indicator to obtain the best material flow state classification model.

[0026] Furthermore, the material flow state classification is obtained by processing the real-time material flow state data through the optimal material flow state classification model, a preset sliding window is used to receive normal material flow state data, and the belt material flow state is updated according to the consistency of the data state in the sliding window, including:

[0027] Generate the material flow state probability based on the real-time single set of distance and signal strength data using the best material flow state classification model;

[0028] If the probability of material or no material state is greater than 90%, the data is recorded as the corresponding state and included in the sliding window;

[0029] If the abnormal probability is greater than 90% or all probabilities are less than or equal to 90%, the data is recorded as abnormal and not included in the sliding window;

[0030] When the sliding window is not fully filled, wait to receive normal data of material flow status until the window is fully filled;

[0031] When the sliding window is full, determine the consistency of the state of all data streams in the sliding window:

[0032] If the material flow status is consistent, the belt material flow status is updated according to the data material flow status in the sliding window;

[0033] If the material flow status is inconsistent, the material flow status update operation will not be performed;

[0034] When the sliding window is full and new data is received, the new data replaces the data that has been in the sliding window the longest.

[0035] Furthermore, the sliding window size needs to be set during initialization before detection, and when the sliding window is not fully filled, a "no material flow" status signal is output by default.

[0036] An intelligent material flow detection system based on ranging radar technology, including a detection module and a data processing module;

[0037] The detection module is arranged at a certain height directly above the belt conveyor, and includes a ranging radar sensor, a power supply system and a heat preservation module;

[0038] The data processing module comprises:

[0039] A data preprocessing unit receives data from the detection module in units of a single set of distance and signal strength;

[0040] The machine learning unit trains, optimizes, and deploys the material flow state classification model based on the belt material flow state dataset through the ML.NET framework;

[0041] The material flow state classification unit obtains data from the data preprocessing unit and transmits the data to the material flow state classification model trained by the ML.NET framework in units of a single set of data to generate the material flow state probability and judge the material flow state of the current data;

[0042] Sliding window unit, establishes a sliding window of preset size to receive normal data of material flow status;

[0043] The material flow status output unit updates the belt material flow status according to the consistency of the data status in the sliding window.

[0044] Furthermore, the data sampling period of the detection module is a customizable value not less than 100ms.

[0045] Since the present invention adopts the above technical solution, the present invention has the following advantages:

[0046] 1. The present invention avoids direct physical contact with the material flow and realizes non-contact detection by adopting a detection module based on a ranging radar sensor. While achieving high-precision detection, it effectively reduces the overall hardware cost and maintenance cost. In addition, the ranging radar sensor has strong anti-interference ability and can operate stably under harsh working conditions, further improving the system reliability and long-term operation stability.

[0047] 2. Traditional material flow detection methods often rely on simple threshold judgments or statistical analysis, which are easily affected by short-term abnormal data or environmental changes, resulting in error accumulation and judgment lags. The present invention combines machine learning technology to effectively improve the system's detection accuracy and real-time performance of material flow status. At the same time, sliding window technology is used to avoid the interference of short-term abnormal data, avoid the occurrence of error accumulation, and ensure the stability and accuracy of the detection results.

[0048] 3. In terms of algorithm selection, the present invention adopts an evaluation index based on macro-precision to select the optimal detection algorithm. The material flow state classification model is obtained through supervised learning based on random dual coordinate ascent (SDCA), limited memory BFGS, lightweight gradient boosting machine (LightGBM), fast decision tree, and fast decision forest algorithm, and a comparative analysis is performed, which can select the most suitable solution from a variety of algorithms; the fusion of different algorithms enables the advantages of each algorithm to be fully utilized, thereby improving the overall classification performance. Compared with a single algorithm, it can effectively improve the detection accuracy of the material flow state classification model and reduce the risk of overfitting. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 Schematic diagram of the flow chart of material flow state classification model training in an embodiment of the present invention.

[0050] Figure 2It is a schematic diagram of the process of the intelligent material flow detection method in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The technical solution of the present invention is further specifically described below by way of embodiments and in conjunction with the accompanying drawings. In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without violating the connotation of the present invention, so the present invention is not limited by the specific embodiments disclosed below. Example

[0052] First, the intelligent material flow detection system of this embodiment is described, which specifically includes:

[0053] Detection module: The detection module is mainly composed of a ranging radar sensor, a power supply system and a thermal insulation module; the detection module is installed at a certain distance above the belt conveyor, and performs non-contact precise measurement through the sensor to collect distance and signal strength data; the data sampling period of the detection module is a customizable value of not less than 100ms; the collected data is transmitted to the data processing module to provide a data basis for subsequent data processing, analysis and material flow status judgment.

[0054] Compared with the prior art, the present invention adopts non-contact detection based on ranging radar technology, which is different from the prior art in sensor selection, and avoids the limitations of the prior art in using sensors for detection under durability, cost, and complex working conditions.

[0055] Data processing module: The hardware of the data processing module is composed of a computer that supports the ML.NET framework and is connected to the detection module. Logically, it includes: a data preprocessing unit, which receives data from the detection module in units of a single set of distance and signal strength; a machine learning unit, which uses the belt material flow state data set to train, optimize and deploy the material flow state classification model based on multiple algorithms such as fast decision trees through the ML.NET framework; a material flow state classification unit, which obtains data from the data preprocessing unit and transmits it in units of a single set of data to the material flow state classification model trained by the ML.NET framework to generate the material flow state probability and judge the current data material flow state; a sliding window unit, which establishes a sliding window with adjustable size to receive normal material flow state data; a material flow state output unit, which updates the belt material flow state according to the consistency of the data state in the sliding window.

[0056] Among them, the sliding window unit is used to receive normal real-time data on material flow status. During the data reception process, when the sliding window is not full, new data will be filled into the window in sequence; once the window is full, the new data will replace the earliest data in the window, so that the latest material flow status data is always retained in the window; the window size can be customized according to actual needs to adapt to the detection accuracy and real-time requirements under different working conditions.

[0057] Compared with the prior art, the present invention adopts a material flow state classification model to judge the material flow state, and adopts a sliding window to ensure that the system can respond correctly according to real-time data, thereby improving the robustness and intelligence level of the material flow detection system.

[0058] The material flow state classification model training method of this embodiment is as follows: Figure 1 As shown, the specific steps include:

[0059] The detection module based on ranging radar technology samples the distance and signal strength data on the upper surface of the belt at a certain time interval (≥100ms) to construct the original data set of the belt material flow status.

[0060] In the data processing module, the original data set of the belt material flow state is preprocessed, and the mean, variance and peak value of the distance and signal strength data are analyzed to filter out the junk data that affects the training of the material flow state classification model, so as to improve the quality of the belt material flow state data set; the belt material flow state data set is divided into training set, validation set and test set in a ratio of 6:3:1, and the training set and validation set data are labeled according to three categories: material, no material and abnormal; in the machine learning unit, the ML.NET framework is used to perform supervised learning based on the random double coordinate ascent (SDCA) algorithm, limited memory BFGS (L-BFGS) algorithm, lightweight gradient boosting machine (LightGBM) algorithm, fast decision tree algorithm and fast decision forest algorithm, and the material flow state classification model is evaluated with macro precision as the evaluation index. The closer the macro precision is to 1, the better the classification performance of the material flow state classification model for all categories. It is used as an evaluation index to obtain the optimal material flow state classification model based on the fast decision tree algorithm, and the optimal material flow state classification model is deployed in the material flow state classification unit.

[0061] The intelligent material flow detection method of this embodiment is as follows: Figure 2 As shown, the specific steps include:

[0062] The system needs to be initialized before detection. At this time, the user needs to customize the sliding window size, and when the window is not full, the default output is the "no material flow" status signal.

[0063] The data collected in real time by the detection module is transmitted to the data preprocessing unit of the data processing module. Then the material flow state classification unit obtains data from the data preprocessing unit and transmits it to the material flow state classification model trained by the ML.NET framework in units of a single set of distance and signal strength data to generate the material flow state probability and judge the material flow state of the current data; if the probability of material or no material state is greater than 90%, the data is recorded as the corresponding state and included in the sliding window; if the abnormal probability is greater than 90% or all probabilities are ≤90%, the data is recorded as an abnormal state and is not included in the sliding window; when the sliding window unit is not full, wait for the data transmitted from the material flow state classification unit until the window is full; when the sliding window unit is full, judge the consistency of the material flow state of all data in the sliding window: if the material flow state is consistent, the material flow state output unit updates the belt material flow state according to the material flow state of the data in the sliding window; if the material flow state is inconsistent, the material flow state update operation is not performed. When the sliding window unit is full and new data is received, the new data will replace the data that has existed in the sliding window for the longest time.

Claims

1. An intelligent material flow detection method based on ranging radar technology, characterized in that: The following steps are involved: The material flow state data of the belt conveyor when it is working is collected in units of a single set of distance and signal strength, and the original data set of the belt material flow state is constructed; Preprocess the original data set of belt material flow status, filter out junk data, obtain the belt material flow status data set, divide the belt material flow status data set into training set, verification set and test set, and annotate the training set and verification set data; The belt material flow state data set is used to train the material flow state classification model through the ML.NET framework, including: using the ML.NET framework to adopt the random dual coordinate ascent algorithm, limited memory BFGS algorithm, lightweight gradient boosting algorithm, fast decision tree algorithm and fast decision forest algorithm to train based on the belt material flow state data set; Calculate the macro accuracy of the material flow state classification model, and use the macro accuracy as a performance indicator to evaluate the material flow state classification model to obtain the optimal material flow state classification model; The material flow state classification is obtained by processing the real-time material flow state data through the optimal material flow state classification model, and the normal material flow state data is received using a preset sliding window, and the belt material flow state is updated according to the consistency of the data state in the sliding window, including: Generate the material flow state probability based on the real-time single set of distance and signal strength data using the best material flow state classification model; If the probability of material or no material state is greater than 90%, the data is recorded as the corresponding state and included in the sliding window; If the abnormal probability is greater than 90% or all probabilities are less than or equal to 90%, the data is recorded as abnormal and not included in the sliding window; When the sliding window is not fully filled, wait to receive normal data of material flow status until the window is fully filled; When the sliding window is full, determine the consistency of the state of all data streams in the sliding window: If the material flow status is consistent, the belt material flow status is updated according to the data material flow status in the sliding window; If the material flow status is inconsistent, the material flow status update operation will not be performed; When the sliding window is full and new data is received, the new data replaces the data that has been in the sliding window the longest.

2. According to claim 1, the intelligent material flow detection method based on ranging radar technology is characterized in that: The preprocessing of the original data set of the belt material flow state includes: analyzing the mean, variance, and peak value based on the distance and signal strength data, filtering out junk data that affects the training of the material flow state classification model, and improving the quality of the belt material flow state data set.

3. The intelligent material flow detection method based on ranging radar technology according to claim 1 is characterized in that: Before the belt material flow state data set is used to train the material flow state classification model through the ML.NET framework, the following steps are included: According to the distance and signal strength characteristics of the belt under different material flow states, the data of the training set and the verification set in the belt material flow state dataset are annotated into three states: with material, without material and abnormal.

4. According to claim 1, the intelligent material flow detection method based on ranging radar technology is characterized in that: The method of calculating the macro accuracy of the material flow state classification model and evaluating the material flow state classification model using the macro accuracy as a performance indicator to obtain the optimal material flow state classification model includes: There are three types of belt material flow states: material, no material, and abnormal. Let A1, A2, and A3 be the accuracy rates of material, no material, and abnormal categories respectively, then the macro accuracy is expressed as: ， The closer the macro accuracy is to 1, the better the classification performance of the material flow state classification model for all categories. It is used as an evaluation optimization indicator to obtain the best material flow state classification model.

5. The intelligent material flow detection method based on ranging radar technology according to any one of claims 1 to 4, characterized in that: The sliding window size needs to be set during initialization before detection, and when the sliding window is not fully filled, the "no flow" status signal is output by default.

6. An intelligent material flow detection system based on ranging radar technology, used to execute the intelligent material flow detection method based on ranging radar technology according to any one of claims 1 to 4, characterized in that: It includes a detection module and a data processing module; The detection module is arranged at a certain height directly above the belt conveyor, and includes a ranging radar sensor, a power supply system and a heat preservation module; The data processing module comprises: A data preprocessing unit receives data from the detection module in units of a single set of distance and signal strength; The machine learning unit trains, optimizes, and deploys the material flow state classification model based on the belt material flow state dataset through the ML.NET framework; The material flow state classification unit obtains data from the data preprocessing unit, transmits the data to the material flow state classification model in units of a single set of data to generate the material flow state probability and judge the material flow state of the current data; Sliding window unit, establishing a sliding window of preset size to receive normal data of material flow status; The material flow status output unit updates the belt material flow status according to the consistency of the data status in the sliding window.

7. The intelligent material flow detection system based on ranging radar technology according to claim 6, characterized in that: The data sampling period of the detection module is a customizable value not less than 100ms.

Images