Coke residue characteristic detection method and system based on visual tactile sensor

Abstract

The invention discloses a coke residue characteristic detection method and system based on a visual tactile sensor, and the method comprises the steps: carrying out the contact scanning of a coke residue sample through the visual tactile sensor, and obtaining the surface three-dimensional topological point cloud data and distribution force vector data of the coke residue sample; carrying out regional or multi-angle scanning on the coke residue sample, and fusing and reconstructing surface three-dimensional topological point cloud data obtained by each scanning into a complete three-dimensional surface model; extracting a plurality of predefined feature parameters from the three-dimensional surface model and the distribution force data, and performing normalization processing to form a structured feature vector; and based on the structured feature vector, coke residue characteristic detection is carried out through a predefined discrimination rule or a machine learning model. According to the method, the visual tactile sensor is used for performing three-dimensional reconstruction and distribution force acquisition on the characteristics of the coke residue, the apparent description of manual visual inspection is converted into quantifiable three-dimensional shape parameters and characteristic indexes, and the problems that manual judgment is high in subjectivity and cannot be quantized and traced are solved.

Description

Technical Field

[0001] This invention relates to the field of coal quality analysis technology, and in particular to a method and system for detecting the characteristics of coke residue based on a visual-tactile sensor. Background Technology

[0002] Char Residue Characteristic (CRC) refers to the morphology of coal samples after volatile matter determination, used to evaluate the caking and coking properties of coal. According to national standards, char residue residues are classified into eight characteristics (1-8) based on shape and degree of adhesion. However, current CRC grading primarily relies on manual visual observation of char residue morphology and finger pressure to determine its adhesion / expansion characteristics. This method has significant subjective errors, heavily influenced by the operator's experience and judgment, leading to poor repeatability. For example, different personnel may have inconsistent perceptions of the "slight adhesion" or "micro-expansion" degree of the same char residue sample, introducing uncertainty into coal quality analysis. Furthermore, manual visual inspection makes it difficult to preserve data from the judgment process, only recording the final grading result, lacking quantitative evidence and traceability of the process. Currently, there is a lack of dedicated automated devices for char residue characteristic determination. Some studies have attempted to analyze char residue images using computer vision, but static images alone cannot fully reflect the mechanical adhesion and internal structural characteristics of char residue samples. Simultaneously, visual methods alone cannot quantify the force feedback obtained from finger pressure. Therefore, it is necessary to develop an automatic detection method that combines tactile sensing and visual reconstruction, which can capture the three-dimensional morphology of coke residue samples and obtain their stress response, so as to more objectively determine the CRC level.

[0003] The "Method and System for Classifying and Detecting Coal Slag Samples" disclosed in Chinese patent literature, publication number CN120354183A, and publication date 2025-07-22, includes: acquiring the noise-reduced spectrum of each slag sample; acquiring the noise index of each wavelength of the spectrum of each slag sample based on the difference between the spectrum of the slag sample and the corresponding noise-reduced spectrum, and the fluctuation frequency of the spectrum of the slag sample; acquiring the structural characteristics of each wavelength based on the difference in noise index among all different slag samples of the same type; acquiring the compositional characteristics of each wavelength based on the absorbance differences of the same wavelength among different slag samples and the absorbance differences of the same wavelength among slag samples of the same type with different degrees of crushing; and acquiring characteristic wavelengths by combining the structural and compositional characteristics of the wavelengths to establish a classification and detection model for slag samples. This technology constructs a classification and detection model for slag samples, improving the accuracy and efficiency of slag sample classification and detection. However, this technology uses near-infrared spectroscopy to classify coke residue by its composition. Similar to techniques that rely solely on static images for coke residue analysis, it is difficult to fully reflect the mechanical adhesion and internal structural characteristics of coke residue samples, and it cannot quantify the force feedback obtained by pressing with a finger, thus failing to objectively determine the CRC level of the coke residue. Summary of the Invention

[0004] This invention aims to overcome the problems of existing coke residue analysis techniques, which are unable to fully reflect the mechanical adhesion and internal structural characteristics of coke residue samples, cannot quantify the force feedback obtained by finger pressure, and thus cannot objectively determine the CRC level of coke residue. It provides a coke residue characteristic detection method and system based on visual-tactile sensors.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A method for detecting the characteristics of coke residue based on a visual-tactile sensor includes: A visual-tactile sensor was used to perform contact scanning on the coke residue sample to obtain three-dimensional topological point cloud data and distributed force vector data of the surface of the coke residue sample. The coke residue sample was scanned in sections or from multiple angles, and the surface three-dimensional topological point cloud data obtained from each scan were fused and reconstructed into a complete three-dimensional surface model. Several predefined feature parameters are extracted from the three-dimensional surface model and distributed force data, and normalized to form a structured feature vector. Based on the structured feature vector, coke residue characteristics are detected using predefined discrimination rules or machine learning models.

[0006] This invention utilizes visual-tactile sensors to perform three-dimensional reconstruction and force distribution acquisition of coke residue properties. It transforms the visual descriptions of cohesiveness, expansion, and other appearances into quantifiable three-dimensional morphological parameters and stress characteristics, ensuring repeatability and consistency of results and providing traceable data support for subsequent judgments. It can capture the three-dimensional morphology of coke residue samples (e.g., volume, height, surface pores, and other signs of expansion) and obtain their stress response (e.g., stress distribution during compression and compressive strength), enabling a more objective determination of the CRC level and solving the problems of subjective, unquantifiable, and untraceable results in manual judgment.

[0007] Preferably, the contact scanning includes: The visual-tactile sensor is pressed into the sampling position of the coke residue sample at a constant speed. When the contact force between the visual-tactile sensor and the coke residue sample reaches a preset threshold, data acquisition is automatically triggered. After the acquisition is completed, the visual-tactile sensor automatically lifts up and detaches from the contact.

[0008] Preferably, the extraction of several predefined feature parameters includes at least one of the following: maximum indentation depth, elastic recovery rate, normal force density distribution, tangential force density distribution, surface expansion height, and reconstructed surface deviation index.

[0009] Preferably, the maximum indentation depth is the difference between the height of the visual-tactile sensor at the initial contact with the coke residue sample at the same position on the surface of the coke residue sample and the lowest height of the visual-tactile sensor under the maximum loading force. The surface expansion height is defined as the height of the highest point of the coke slag sample, with the bottom plane of the coke slag sample as a reference.

[0010] Preferably, the elastic recovery rate is: the difference between the maximum indentation depth generated by the visual-tactile sensor at the same location on the surface of the coke residue sample and the residual deformation depth after unloading, divided by the percentage of the maximum indentation depth.

[0011] Preferably, the normal force density distribution is: the force density obtained by dividing the normal force at each sampling position of the coke residue sample by the corresponding contact area; The tangential force density distribution is: the force density obtained by dividing the tangential force at each sampling location of the coke residue sample by the corresponding contact area.

[0012] Preferably, the reconstructed surface deviation index is: the standard deviation of the height of each point cloud in the three-dimensional surface model of the coke slag sample; or The ratio of the actual surface area of ​​the coke residue sample to the corresponding projected area of ​​the convex hull.

[0013] Preferably, a feedback correction step is also included: After obtaining the detection results of the coke residue characteristics, the judgment level of the detection results is verified and corrected through the human-computer interaction interface, and the judgment rules or machine learning models are updated based on the corrected data.

[0014] Preferably, the process of fusing and reconstructing the surface three-dimensional topological point cloud data obtained from each scan into a complete three-dimensional surface model includes: An iterative nearest point or feature alignment algorithm is used to align the surface 3D topological point cloud data obtained from scanning adjacent sampling positions in a unified coordinate system and fuse them into a complete 3D surface model; and a mesh model is generated through a surface reconstruction algorithm.

[0015] A coke residue characteristic detection system based on a visual-tactile sensor includes: The motion control module, including a three-axis gantry slide and a lead screw stepper motor, controls the precise positioning of the visual and tactile sensors; The sensor module includes a visual and tactile sensor, which is fixed to the bottom of the Z-axis slide via a flange to collect three-dimensional morphology and force distribution data of the coke residue sample; the Z-axis slide controls the vertical lifting and lowering of the sensor. The sample positioning module, including a measuring stage and a vision positioning module, fixes the coke residue sample and performs coarse positioning. The control and data processing module, including edge computing devices, performs motion control, data acquisition, 3D reconstruction, feature extraction, and discrimination logic.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention utilizes point cloud and mechanical data acquired by a high-precision tactile vision sensor to replace subjective human observation, enabling the assessment of coke residue characteristics to be based on objective measurement. Properties such as adhesion and expansion are no longer merely qualitative sensory descriptions, but are reflected through quantifiable indicators such as maximum indentation depth, surface expansion height, and peak stress, greatly improving the scientific rigor and credibility of the results.

[0017] 2. The system of this invention fully records the original three-dimensional data, mechanical response time sequence, and finally extracted feature parameters for each scan. Data is retained throughout the judgment process for each coke residue sample, allowing for easy review of the original point cloud morphology, stress distribution image, and judgment criteria. This provides a basis for quality control and dispute resolution, meeting the data traceability requirements in laboratory management.

[0018] 3. This invention, through standardized sensor calibration and scanning procedures, ensures good consistency in measurement results for the same sample across different batches and devices, significantly reducing fluctuations caused by human factors. The system automatically triggers data acquisition and a unified data processing algorithm, guaranteeing consistent experimental conditions and thus ensuring high repeatability of results.

[0019] 4. The system of this invention can be integrated with an automatic sample feeding device and control system to automate the coke residue identification process. Sample scanning, data analysis, and grade determination can be completed without manual intervention, significantly improving efficiency and reducing labor costs. For coal quality testing laboratories, the system of this invention can be deployed as unattended equipment to upgrade existing coke residue characteristic identification processes and greatly enhance the level of intelligence in testing. Attached Figure Description

[0020] Figure 1 This is a flowchart of the coke residue characteristic detection method based on visual-tactile sensors in this invention.

[0021] Figure 2 This is a schematic diagram of the coke residue characteristic detection system based on a visual-tactile sensor in this invention.

[0022] Figure 3 This is the visualization interface of the three-dimensional point cloud model of coke slag in this embodiment of the invention.

[0023] Figure 4 This is the visualization interface of the grid model of coke slag in the embodiments of the present invention.

[0024] Figure 5 This is a visualization interface of the deformation field diagram of coke slag in the embodiments of the present invention.

[0025] Figure 6 This is a visualization interface of the distribution force diagram of coke residue in the embodiments of the present invention.

[0026] In the figure: 1. Edge computing device; 2. Industrial camera; 3. Coke residue sample; 4. Slide table; 5. Vertical lead screw stepper motor; 6. Visual and tactile sensor; 7. Planar lead screw stepper motor. Detailed Implementation

[0027] The present invention will now be further described with reference to the accompanying drawings and specific embodiments.

[0028] like Figure 1 As shown, a method for detecting the characteristics of coke residue based on a visual-tactile sensor includes: A visual-tactile sensor was used to perform contact scanning on the coke residue sample to obtain three-dimensional topological point cloud data and distributed force vector data of the surface of the coke residue sample. The coke residue sample was scanned in sections or from multiple angles, and the surface three-dimensional topological point cloud data obtained from each scan were fused and reconstructed into a complete three-dimensional surface model. Several predefined feature parameters are extracted from the three-dimensional surface model and distributed force data, and normalized to form a structured feature vector. Based on the structured feature vector, coke residue characteristics are detected using predefined discrimination rules or machine learning models.

[0029] This invention utilizes visual-tactile sensors to perform three-dimensional reconstruction and force distribution acquisition of coke residue properties. It transforms the visual descriptions of cohesiveness, expansion, and other appearances into quantifiable three-dimensional morphological parameters and stress characteristics, ensuring repeatability and consistency of results and providing traceable data support for subsequent judgments. It can capture the three-dimensional morphology of coke residue samples (e.g., volume, height, surface pores, and other signs of expansion) and obtain their stress response (e.g., stress distribution during compression and compressive strength), enabling a more objective determination of the CRC level and solving the problems of subjective, unquantifiable, and untraceable results in manual judgment.

[0030] With the development of sensing technology, a new type of sensor integrating computer vision and tactile perception has emerged—the visual-tactile sensor. A visual-tactile sensor is a high-resolution tactile sensing device based on vision, capable of contacting the measured object through a soft medium within a compact space, and sensing information such as the three-dimensional surface morphology and force distribution of the contact area in real time. This type of sensor uses embedded optical measurement methods to reconstruct the deformation of the medium in 3D, thereby acquiring rich data such as the point cloud and normal / tangential force distribution of the contact surface. Visual-tactile sensors can accurately measure multimodal tactile information such as contact morphology, three-dimensional deformation field, contact distribution force, and six-dimensional concentrated force. The emergence of this sensor provides the hardware foundation for the automation of coal slag characteristic analysis, converting the appearance characteristics observed manually into quantitatively extractable structured data. By analyzing the residual slag after coal volatile matter determination, it replaces the manual visual inspection and experience-based judgment in the traditional slag characteristic (CRC) category 1-8 determination process, achieving objective, quantitative, and traceable acquisition of the appearance morphology and mechanical characteristics of slag.

[0031] Furthermore, after completing the initial coke residue characteristic testing, a feedback correction step is also included: After obtaining the detection results of the coke residue characteristics, the judgment level of the detection results is verified and corrected through the human-computer interaction interface, and the judgment rules or machine learning models are updated based on the corrected data.

[0032] Specifically, after the automatic system provides preliminary detection results of coke slag characteristics, coal quality analysis experts can verify and provide feedback on the results by viewing the 3D model and stress distribution visualization of the coke slag sample. If the detection results do not match the expert's experience, the expert can mark a correction and incorporate this feedback into the system database. Based on this, the expert can adjust the discrimination rules or retrain the machine learning model to continuously optimize the accuracy of the judgment. This human-computer interaction mechanism ensures that the system using the method of this invention can initially combine expert knowledge to improve the reliability of the detection results, while simultaneously achieving continuous adaptive improvement of the model. Experts can also adjust scanning parameters, feature extraction methods, or judgment thresholds to adapt to special coal types or abnormal samples, giving the system a certain degree of flexibility and adjustability.

[0033] As a specific embodiment, the process of contact scanning of the coke residue sample using a visual-tactile sensor in this embodiment includes: The visual-tactile sensor is pressed into the sampling position of the coke residue sample at a constant speed. When the contact force between the visual-tactile sensor and the coke residue sample reaches a preset threshold, data acquisition is automatically triggered. After the acquisition is completed, the visual-tactile sensor automatically lifts up and detaches from the contact.

[0034] Specifically, visual-tactile sensing is used to perform contact scanning on coke residue samples retained in the crucible after volatile matter determination. The flexible tactile sensing surface of the sensor makes point-by-point contact with the coke residue surface, acquiring high-resolution three-dimensional topological point cloud data of the surface, as well as force distribution vector data (including normal and tangential components) generated during the contact process. During the scanning process, array-type optical acquisition records the changes in the deformation field of the medium, realizing real-time reconstruction of the microstructure of the contact surface and sensing of the contact mechanical response.

[0035] The data acquisition method is set to automatic triggering: when the sensor touch surface detects contact with the sample and the contact force reaches a preset threshold and stabilizes, the point cloud and distributed force data of the current area are immediately recorded. Then, the visual tactile sensor detaches from the coke sample surface and moves to the next predetermined position for scanning. Throughout the scanning process, the contact force and speed are uniformly controlled to avoid human error.

[0036] Furthermore, since the visual-tactile sensor can only collect data from a portion of the coke residue sample at a time, in order to obtain complete data from the coke residue sample, multiple data collections are performed on different areas of the coke residue sample surface. Then, the surface three-dimensional topological point cloud data obtained from each scan are fused and reconstructed into a complete three-dimensional surface model.

[0037] The fusion and reconstruction process includes: using an iterative nearest point or feature alignment algorithm to align the surface 3D topological point cloud data obtained from scanning adjacent sampling positions in a unified coordinate system and fuse them into a complete 3D surface model; and generating a mesh model through a surface reconstruction algorithm.

[0038] Specifically, to address the limited coverage of a single scan, the coke residue sample is scanned multiple times in different regions, or multi-angle scans are performed by changing the relative angle between the sensor and the sample. Point cloud stitching and registration algorithms are used to align the point cloud data obtained from each scan in a unified coordinate system, forming a complete three-dimensional surface model of the coke residue sample. The outline and surface details of the coke residue can be presented in the form of point clouds or triangular meshes and visualized. For dynamic observation, a series of point clouds can be acquired at different contact moments to achieve temporal reconstruction of the coke residue surface morphology, force field, and deformation field, used to analyze the gradual deformation and mechanical changes during contact loading.

[0039] As a specific implementation, to quantify the physical properties of coke slag, a series of key statistics and characteristic parameters are extracted from the three-dimensional morphology and distributed force data of the coke slag sample. Several predefined characteristic parameters are extracted, including at least one of the following: maximum indentation depth, elastic recovery rate, normal force density distribution, tangential force density distribution, surface expansion height, and reconstructed surface deviation index. All extracted characteristic parameters are stored in a structured manner to provide a reliable quantitative basis for subsequent analysis and model discrimination.

[0040] Maximum indentation depth: Reflects the maximum deformation depth of coke residue under pressure, and is used to characterize its hardness and compressibility.

[0041] Its quantitative definition is the maximum displacement distance of the sensor's flexible touch surface pressed into the coke residue sample from the initial contact position. For example, if the initial contact height of the visual-tactile sensor is H0, and the minimum height under the maximum loading force is H1, then the maximum indentation depth h is... max h is the difference between the initial contact height H0 of the visual-tactile sensor with the coke slag sample at the same location on the sample surface and the minimum height H1 of the visual-tactile sensor under maximum loading force (unit: mm). This maximum value can be directly recorded in a single scan; if different areas of the sample are pressed into contact multiple times, the extreme value (maximum value) of each indentation depth can be taken as the overall h of the sample. max The mean and standard deviation of the indentation depth distribution are calculated as needed to assess the uniformity of coke hardness.

[0042] Local elastic recovery rate: characterizes the degree of recovery of surface deformation after pressure is released, and is used to evaluate the elastic-plastic behavior of coke residue, such as the rebound ability of bonded coke.

[0043] The local elastic recovery rate η can be calculated by the difference in morphology before and after indentation: the maximum indentation depth h generated by the visual-tactile sensor at the same location on the surface of the coke residue sample. max Residual deformation depth h after unloading res The difference between (i.e., the depth of the permanent indentation on the coke slag surface) and the maximum indentation depth h is then divided. max The percentage results are given. η=0% indicates completely irreversible plastic deformation, and η=100% indicates complete elastic recovery with no residual deformation. In actual testing, the surface morphology at that location can be re-collected after a certain period of time following the removal of the maximum load to measure h. res If multiple measurements are taken, the average recovery rate can be taken and the dispersion recorded to characterize the consistency of elastic recovery in different parts.

[0044] Surface expansion height H eThis refers to the vertical height of the coke residue sample relative to a reference plane, used to quantify the degree of expansion of the coke residue. The height H of the highest point of the coke residue is measured, using the bottom plane of the coke residue residue as a reference (e.g., the bottom of the crucible or the lowest point of the coke residue as a reference surface). max (Unit: mm). The distance between the highest point and the reference plane is defined as the expansion height H. e =H max For irregularly shaped coke residue, its outline can be obtained through reconstructed 3D point cloud, and the coordinate difference between the highest point and the reference plane can be directly calculated to obtain H. e Expansion height is an important indicator for determining the expansion and melting characteristics of coke slag; generally, a greater height indicates more pronounced expansion. For example, according to national standards, coke slag with an expansion height exceeding 15 mm is typically classified as exhibiting strong expansion characteristics. In statistical analysis, the expansion height of samples and their distribution among different samples can be recorded for comparison with manual rating standards.

[0045] Normal / tangential force density distribution: describes the distribution density of normal pressure and tangential friction on the contact surface, reflecting surface roughness (corresponding to fluctuations in normal pressure) and adhesion characteristics (corresponding to the magnitude of tangential friction).

[0046] The sensor's embedded vision measurement can acquire the normal force at each sampling point on the surface of the coke residue sample. and tangential force Divide the force by the corresponding contact area element. Force density (stress) is obtained. Normal force density distribution. The force density is calculated by dividing the normal force at each sampling location of the coke residue sample by the corresponding contact area; the tangential force density distribution is also described. The force density is calculated by dividing the tangential force at each sampling location of the coke residue sample by the corresponding contact area, expressed in Pascals (Pa). This results in distribution cloud maps of the normal and tangential force density distributions.

[0047] Peak stress location and distribution characteristics: Identifying the location and distribution of the maximum stress during the contact process helps to determine the uniformity and potential weaknesses of the internal structure of coke residue.

[0048] Statistical features, such as the peak value of the normal pressure, can be further extracted from the normal / tangential force density distribution. and its coordinates Peak value of tangential stress and the mean of the pressure distribution. and standard deviation These quantities can quantitatively describe the characteristics of force distribution: for example... and The ratio reflects the degree of pressure concentration, and the standard deviation The overall level of the tangential force distribution reflects the uniformity of the force distribution and the magnitude of the frictional resistance of the coke slag surface to the sensor.

[0049] Reconstruction surface deviation index: used to quantify the degree of surface undulation and shape deviation of coke residue, and can indicate the degree of expansion and bubbling or morphological complexity of coke residue.

[0050] The calculation of the reconstructed surface deviation index is based on the reconstructed three-dimensional topographic data. One approach is to calculate a statistical measure of the height deviation of the coke slag surface relative to a reference plane, such as the standard deviation of the surface height. : Point cloud z in the three-dimensional surface model of the coke slag sample i The standard deviation of the height, where z i Let be the height of the i-th point in the surface point cloud. A larger value indicates more pronounced surface undulations and a more significant deviation of the morphology from the average plane. Another approach is to compare the actual surface area S of the coke residue with the projected area of ​​its convex hull. The ratio R is a factor indicating that the surface unevenness area is larger than the projected area, suggesting the presence of more pores, bulges, and other complex structures. The above deviation indices can be appropriately defined based on specific needs, and auxiliary parameters such as the maximum deviation value (e.g., the height difference between the highest and lowest peaks) can be extracted. Statistical analysis of a series of deviation parameters can quantify the roughness and complexity of the coke slag surface, providing a basis for determining the expansion characteristics of the coke slag.

[0051] The extracted multidimensional feature parameters are organized into structured feature vectors or tables for auxiliary discrimination of coke residue feature levels (CRC 1-8). Empirical rules or machine learning models (e.g., simple decision tree classifiers or convolutional neural network-based pattern recognition models) can be pre-established based on a large number of historical samples to utilize these features for classification. This embodiment provides a standardized data interface, allowing the feature output to be easily integrated into subsequent detection modules for automatic judgment or decision support. When using a rule-based method, thresholds or combinations of judgment criteria for each feature parameter can be set to correspond to different levels in national standards. When using a machine learning model, training can enable the model to learn the mapping relationship between feature parameters and CRC levels, thereby providing a predicted level when new sample features are input.

[0052] As a specific embodiment, in addition to providing a method for detecting the characteristics of coke residue based on a visual-tactile sensor, the present invention also provides, as a... Figure 2 The system shown is a coke residue characteristic detection system based on visual and tactile sensors, including: a motion control module, a sensor module, a sample positioning module, and a control and data processing module.

[0053] The motion control module, including a three-axis gantry slide and a lead screw stepper motor, is used to control the precise positioning of the visual tactile sensor in the X, Y, and Z axes.

[0054] The three-axis gantry slide structure includes corresponding slides 4. The X-axis slide controls the horizontal movement (forward and backward) of the sensor; the Y-axis slide controls the horizontal movement (left and right) of the sensor; and the Z-axis slide controls the vertical lifting of the sensor. Each axis is equipped with an independent lead screw stepper motor for precise positioning. Planar lead screw stepper motors 7 are used for the X-axis and Y-axis slides, and a vertical lead screw stepper motor 5 is used for the Z-axis slide. In the high-precision lead screw drive system of the lead screw stepper motor, a ball screw pair with a lead of 5mm is used to achieve a positioning accuracy of 0.01mm. The stepper motor, in conjunction with a microstepping driver, achieves a typical resolution of 1600 steps / revolution. The effective stroke is: X-axis ≥300mm, Y-axis ≥200mm, and Z-axis ≥150mm.

[0055] The sensor module includes a visual-tactile sensor 6, which is fixed to the bottom of the Z-axis slide via a flange. The flexible tactile sensing surface of the sensor faces vertically downwards and can contact the surface of the coke residue sample 3. The visual-tactile sensor 6 incorporates a high-resolution industrial camera (for real-time acquisition of images of tactile medium deformation) and a six-dimensional force sensor (for real-time monitoring of contact force), thereby acquiring three-dimensional morphology and force distribution data of the coke residue sample.

[0056] The sensor module also includes a force feedback unit. When the detected contact force reaches a preset threshold (e.g., 5N), it automatically triggers data acquisition. The Z-axis stepper screw motor immediately stops descending and maintains a constant pressing depth. After the data acquisition is completed, the Z-axis stepper screw motor moves in the opposite direction, lifting the sensor and disengaging it from the coke sample.

[0057] The sample positioning module includes a measuring stage and a vision positioning module. The measuring stage is used to fix the coke residue sample 3 or crucible, and can be equipped with a rotating stage to realize 360° rotation scanning of the sample. The surface flatness of the stage is ≤0.05mm to ensure the consistency of the reference.

[0058] The visual positioning module achieves coarse positioning of the sample through the top-mounted industrial camera 2, determines the approximate outline and center position of the coke residue through image recognition, automatically generates the starting point coordinates of the scanning path planning, and completes the scanning path planning.

[0059] The control and data processing module, including edge computing device 1, performs motion control, data acquisition, 3D reconstruction, feature extraction and discrimination logic, and outputs visualization results.

[0060] It employs a microcomputer or industrial PC (such as the NVIDIA Jetson series) running a real-time Linux system, responsible for the parallel scheduling of motion control, data acquisition, and image processing. It communicates with actuators and sensors via serial port / Ethernet to achieve efficient data flow management.

[0061] As a specific example, according to national standards, coke residue can be classified into eight characteristics based on shape and degree of adhesion: from type 1, which is entirely powder (particles not adhering to each other), to type 8, which exhibits strong expansion and fusion adhesion (forming expanded blocks with a height exceeding 15 mm and a metallic luster), and so on. It is generally believed that type 1 and 2 coke residues are non-adhesive, type 3 and 4 are weakly adhesive, and type 5 to 8 are strongly adhesive.

[0062] After determining the volatile matter content of a coking coal sample using the national standard method, some coke residue remained in the crucible. Expert experience indicated that the coke residue exhibited characteristic number 6 (molten agglomeration and expansion). This invention's system is used to automatically identify this coke residue. Figure 3-6 The image shown is a visualization of the results of coke residue sample testing using this method. It includes: a reconstructed 3D point cloud / mesh model of the coke residue sample, clearly showing the shape and surface details; a normal force distribution map of the coke residue surface, using color or arrow vectors to represent the pressure magnitude at different locations; and a 3D deformation field during the coke residue contact process, demonstrating the spatial distribution of deformation under pressure.

[0063] The specific implementation steps for detecting coke residue using the method of this invention are as follows.

[0064] Equipment Preparation and Calibration: Mount the visual-tactile sensor on a three-axis motion platform (or the end effector of a robotic arm), ensuring its sensing surface faces downwards and can vertically contact the sample. Perform initial calibration in the visual-tactile sensor software system according to the sensor's usage specifications, including coordinate system calibration, sensor zero-point calibration, and camera imaging parameter correction. Calibration establishes an accurate correspondence between the data acquired by the sensor and the actual scale, and corrects for factors such as ambient light and temperature drift, ensuring the accuracy and reliability of subsequent measurement data. Then, fix the crucible containing coke residue (or place the removed coke sample on the measuring stage) at an appropriate position below the sensor, ready to begin scanning.

[0065] Standardized Scanning Process: A pre-defined scanning process script is invoked. Based on the approximate size and shape of the coke residue sample, the scanning process plans several contact acquisition points or paths. For example, for blocky coke residue, multiple contact points can be planned in the middle and at the edges; for flat, sheet-like coke residue, a grid-like contact scanning path is planned to ensure coverage of the entire sample surface morphology. The system is set to automatically trigger data acquisition. In this embodiment, the sensor is set to descend at a constant speed. When the detected contact force reaches 5N, data is recorded and the sensor is raised, with consistent indentation force for each contact. Through the standardized process, data obtained from multiple scans are comparable and consistent. For samples requiring multi-angle scanning, the sample is rotated or the sensor angle is adjusted according to a predetermined sequence, and the above steps are repeated to obtain data from different perspectives.

[0066] 3D Reconstruction: After all scans are completed, the point cloud data obtained from multiple contact points are fused and reconstructed. First, a filtering algorithm is applied to remove outliers and noise. Then, an ICP (Iterative Nearest Point) or feature alignment algorithm is used to register the point clouds of adjacent scans together. Combined with sensor pose recording, the stitching accuracy is further improved. Finally, a complete 3D point cloud model of the coke slag sample is generated, and the point cloud is converted into a mesh model through surface reconstruction algorithms (such as Delaunay triangulation) to more intuitively display the surface shape and volume of the coke slag.

[0067] For the coke residue in this embodiment (assuming it consists of several pieces of expanded coke that have been bonded together), such as Figure 3 , 4 As shown, the reconstructed 3D model clearly displays the overall blocky morphology of the coke residue. The upper surface is raised with several expanded bubble-like protrusions, while the lower surface is relatively flat and has a metallic luster (visually apparent through different reflectivities, but this sensor primarily focuses on morphology and mechanical data). The maximum height of the reconstructed model is approximately 12mm, and its lateral dimensions are approximately 30mm × 20mm.

[0068] Force and Deformation Field Acquisition: At each contact scanning point, the sensor not only acquires the topographic point cloud but also records the local three-dimensional deformation field and force distribution of the elastic medium. The force distribution of each contact is mapped to the corresponding location in the reconstructed 3D model to generate a mechanical distribution map of the coke surface. In this embodiment, as... Figure 5 , 6As shown, a thermogram of the normal pressure exerted on different regions of the coke residue surface can be obtained: the results show that the central region of the coke residue only produces an indentation of about 0.5 mm under a 5 N indentation force, while some weaker areas at the edges produce indentations of more than 1 mm, accompanied by local fragmentation (manifested as local collapse in the subsequent point cloud). Sensor deformation field data also reveals minute slippage and frictional behavior on the coke residue surface during contact. For example, during one contact, a slight relative slippage occurred between the probe and the coke residue surface, corresponding to a peak in the tangential force field in that region. All this distributed force and deformation information is associated with and stored in the 3D model, constituting a spatiotemporal dataset of the force response of the coke residue sample.

[0069] Feature parameter calculation: Based on the complete 3D model and mechanical data, a series of predefined feature parameters are automatically calculated, including maximum indentation depth, elastic recovery rate, normal force density distribution, tangential force density distribution, surface expansion height, and reconstructed surface deviation index.

[0070] For the coke slag sample in this embodiment, the following were calculated: the maximum indentation depth was 1.2 mm (occurring in the edge region of the coke slag, indicating that the structure there is relatively loose); the local elastic recovery rate was 80% (indicating that most of the deformation recovered after unloading, but there was 20% residual deformation, reflecting that the coke slag has a certain degree of plasticity); the peak force was mainly distributed in the upper part of the middle of the coke slag, corresponding to the position of the maximum value of the sensor's normal force; the normal force vector density was relatively uniformly distributed on the surface of the coke slag, but the tangential force increased significantly at the edge contact, indicating that those areas had a higher coefficient of friction or a rougher surface; the reconstructed surface deviation index was calculated to be 0.15 (a quantitative value of surface unevenness based on an ideal flat plate, indicating that the upper surface of the coke slag has a moderate degree of undulation and expansion). These characteristic parameters are closely related to the grade of coke slag: for example, relatively large indentation depth and residual deformation often correspond to lower adhesion, while the indentation of this sample is small and the elastic recovery is good, which is consistent with the expectation of its high adhesion.

[0071] After obtaining the original feature parameters, normalization is required to convert indicators of different dimensions and orders of magnitude into comparable dimensionless values ​​and standardize them into structured input feature vectors. The purpose of normalization is to eliminate the influence of differences in units or ranges among the indicators, ensuring that the subsequent classification model treats each feature equally. Commonly used normalization methods include: Z-score standardization (zero-mean standardization): Suitable for data that approximates a normal distribution or where it's necessary to preserve the influence of outliers. After Z-score transformation, the new features have a distribution with a mean of 0 and a standard deviation of 1, and no units. This transforms values ​​such as "maximum indentation depth" (in mm) and "normal force" (in N) to a directly comparable scale, preventing features with large absolute values ​​from dominating the model. Z-score standardization also reduces the adverse effects of outliers on the model while preserving the relative differences between samples.

[0072] Min-Max normalization (min-max scaling): Suitable for scaling all feature values ​​to a uniform interval (e.g., [0,1]) to facilitate distance measurement or intuitive comparison of results. After Min-Max normalization, feature values ​​are linearly mapped to the range of 0 to 1: samples with the minimum value are mapped to 0, samples with the maximum value are mapped to 1, and the rest are mapped proportionally to the interval. This method is intuitive and easy to understand, and it preserves the relative magnitudes of the original distribution.

[0073] Logarithmic scaling: When the distribution of certain features is highly skewed or spans multiple orders of magnitude, logarithmic scaling can be used to compress their dynamic range and enhance numerical stability. Taking the logarithm of feature y, such as y' = log(1+y) (when y≥0) or y' = log(y) (when y>0), compresses larger values ​​while widening the relative differences between smaller values. Logarithmic scaling (or other nonlinear scaling such as square root) is often used in conjunction with the above standardization methods: first, take the logarithm of the feature to approximate a symmetrical distribution, then apply Z-score standardization to obtain features whose distribution is closer to normal and whose outlier effects are reduced. It is important to choose an appropriate base (commonly 10 or the natural logarithm) and to perform translation processing on the 0-value special case.

[0074] After appropriate normalization, the multidimensional features of each sample can be represented as a structured feature vector v = [f1, f2, ..., fn], where fi is the standardized value of the i-th feature. This feature vector will serve as input to subsequent machine learning models or discrimination rules. Specifically, a suitable normalization scheme can be selected based on the type of machine learning model: for models optimized based on distance metrics or gradient descent (such as KNN, SVM, and neural networks), Z-score normalization is generally more suitable, as it can accelerate convergence and improve stability; for cases where intuitive interpretation of output probabilities is crucial, Min-Max normalization can be considered to limit the results to the range of 0 to 1 for easier interpretation. All samples should use the same normalization parameters (e.g., the mean and standard deviation should be calculated based on the training set) to avoid data leakage. Finally, the structured feature vector set will be fed into a pre-established machine learning model or discrimination rule to automatically determine the coke residue feature level (CRC 1-8).

[0075] Judgment and Result Output: The above-mentioned structured features are input into the pre-constructed discrimination logic. Assume that the system in this embodiment has a built-in rule-based CRC judgment table: it specifies that if the coke slag height exceeds 15mm and surface expansion bubbles are obvious, it is grade 8; 10-15mm is grade 7; if it has a metallic luster and forms a blocky shape without powder, it is at least grade 5 or higher; the grade is further subdivided by combining other mechanical characteristic thresholds. Based on the calculated characteristics, this sample has a height of 12mm, expansion bubbles, intact coke blocks that are not easily crushed, and high elastic recovery; therefore, it is judged to be grade 7, "expansion-melting-bonding" coke slag. The judgment result is output to the visualization interface, including "Judgment Grade: CRC-7" and a list of key parameter values ​​on which this conclusion is based. Simultaneously, a 3D model of the sample and a visualization of the force distribution are also output for the user to view.

[0076] Expert Verification and Learning Feedback: Analysts review the detection results and compare them with their own experience. If the judgment is deemed reasonable, the result is archived, and the entire set of data (model, features, and levels) is entered into the database to enrich the training sample library. If a deviation in judgment is found, for example, if an expert believes it should be judged as number 6 instead of number 7, the judgment level can be changed in the system, and the reason can be marked (e.g., "the inflation height is slightly lower than the typical value of standard number 7"). The system records this human correction and uses the data of this sample to update the judgment rule model or machine learning model: for rule-based models, the threshold can be adjusted or rules can be added appropriately; for machine learning models, this sample can be used as new training data to correct model parameters. Through multiple iterations, the system's automatic judgment capability will continuously improve, becoming closer to the consistent judgment of experts.

[0077] This invention provides a complete understanding of the morphology and mechanical characteristics of coke slag, digitizing and automating the traditional process of visual inspection and pressure testing, significantly improving the objectivity and reliability of the results. Furthermore, detailed data records are generated for each sample's analysis, facilitating subsequent investigation and analysis. In practical laboratory applications, this method can shorten the time required for coke slag characteristic determination and improve work efficiency. In coal quality monitoring, it can serve as an auxiliary basis for evaluating coal caking properties, providing quantitative references for coking coal blending and process adjustments.

[0078] The above embodiments are further elaborations and descriptions of the present invention to facilitate understanding, and are not intended to limit the present invention in any way. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for detecting the characteristics of coke residue based on a visual-tactile sensor, characterized in that, include: A visual-tactile sensor was used to perform contact scanning on the coke residue sample to obtain three-dimensional topological point cloud data and distributed force vector data of the surface of the coke residue sample. The coke residue sample was scanned in sections or from multiple angles, and the surface three-dimensional topological point cloud data obtained from each scan were fused and reconstructed into a complete three-dimensional surface model. Several predefined feature parameters are extracted from the three-dimensional surface model and distributed force data, and normalized to form a structured feature vector. Based on the structured feature vector, coke residue characteristics are detected using predefined discrimination rules or machine learning models.

2. The method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 1, characterized in that, The contact scanning includes: The visual-tactile sensor is pressed into the sampling position of the coke residue sample at a constant speed. When the contact force between the visual-tactile sensor and the coke residue sample reaches a preset threshold, data acquisition is automatically triggered. After the acquisition is completed, the visual-tactile sensor automatically lifts up and detaches from the contact.

3. A method for detecting the characteristics of coke residue based on a visual-tactile sensor according to claim 1 or 2, characterized in that, The extracted predefined feature parameters include at least one of the following: maximum indentation depth, elastic recovery rate, normal force density distribution, tangential force density distribution, surface expansion height, and reconstructed surface deviation index.

4. The method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 3, characterized in that, The maximum indentation depth is the difference between the height of the visual-tactile sensor at the initial contact with the coke slag sample at the same position on the surface of the coke slag sample and the lowest height of the visual-tactile sensor under the maximum loading force. The surface expansion height is defined as the height of the highest point of the coke slag sample, with the bottom plane of the coke slag sample as a reference.

5. The method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 3, characterized in that, The elastic recovery rate is: the difference between the maximum indentation depth generated by the visual-tactile sensor at the same location on the surface of the coke residue sample and the residual deformation depth after unloading, divided by the percentage of the maximum indentation depth.

6. The method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 3, characterized in that, The normal force density distribution is: the force density obtained by dividing the normal force at each sampling position of the coke residue sample by the corresponding contact area. The tangential force density distribution is: the force density obtained by dividing the tangential force at each sampling location of the coke residue sample by the corresponding contact area.

7. The method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 3, characterized in that, The reconstructed surface deviation index is: the standard deviation of the height of each point cloud in the three-dimensional surface model of the coke slag sample; or The ratio of the actual surface area of ​​the coke residue sample to the corresponding projected area of ​​the convex hull.

8. A method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 1, 2, 4, 5, 6, or 7, characterized in that, It also includes a feedback correction step: After obtaining the detection results of the coke residue characteristics, the judgment level of the detection results is verified and corrected through the human-computer interaction interface, and the judgment rules or machine learning models are updated based on the corrected data.

9. The method for detecting coke residue characteristics based on a visual-tactile sensor according to claim 8, characterized in that, The process of fusing and reconstructing the surface three-dimensional topological point cloud data obtained from each scan into a complete three-dimensional surface model includes: An iterative nearest point or feature alignment algorithm is used to align the surface 3D topological point cloud data obtained from scanning adjacent sampling positions in a unified coordinate system and fuse them into a complete 3D surface model; and a mesh model is generated through a surface reconstruction algorithm.

10. A coke residue characteristic detection system based on a visual-tactile sensor, applicable to the coke residue characteristic detection method as described in any one of claims 1-9, characterized in that, include: The motion control module, including a three-axis gantry slide and a lead screw stepper motor, controls the precise positioning of the visual and tactile sensors; The sensor module includes a visual and tactile sensor, which is fixed to the bottom of the Z-axis slide via a flange to collect three-dimensional morphology and force distribution data of the coke residue sample; the Z-axis slide controls the vertical lifting and lowering of the sensor. The sample positioning module, including a measuring stage and a vision positioning module, fixes the coke residue sample and performs coarse positioning. The control and data processing module, including edge computing devices, performs motion control, data acquisition, 3D reconstruction, feature extraction, and discrimination logic.

Images