Method of adjusting a fiberizing apparatus

Abstract

A method is provided for adjusting at least one fiberization parameter of a fiberization apparatus used in mineral wool manufacturing. The method includes: generating shape information of the centrifuge disc of the fiberization apparatus by imaging a mark on a centrifuge disc and mapping the position of the imaged mark to its corresponding physical position using image analysis; quantifying physical melt parameters relative to the centrifuge disc by analyzing the imaged centrifuge disc on which melt to be fiberized is falling, based on the shape information and using a computer vision system; and calculating fiberization parameters and their target values ​​based on the melt parameters, the fiberization parameters depending on the melt parameters, and adjusting the fiberization apparatus to reduce the deviation from the target value if the fiberization parameters deviate from the target value.

Description

Technical Field

[0001] This invention relates to measuring one or more melt parameters in a fiberization apparatus used for manufacturing mineral wool, and to modifying one or more fiberization parameters by adjusting said apparatus. Typically, the measured melt parameters are the impact positions of the melt on a centrifugal disc, and the apparatus is adjusted to change these positions. Background Technology

[0002] Mineral wool can be manufactured using a variety of different techniques. Typically, a suitable material (such as various types of rock) is heated in a furnace until it melts. This produces a mineral melt, which is then centrifuged into fibers using a fiberizing device. The fiberizing device comprises a series of centrifugal discs, each rotating in the opposite direction to the preceding disc in the series, and arranged such that the melt poured onto the top disc is sequentially thrown onto the surface of each subsequent disc, or thrown away from the series of discs. The resulting fibers are collected as wool.

[0003] The performance of the manufacturing process will vary with many factors, such as the melt impact angle formed between the vertical diameter axis of the centrifugal disc and the impact position of the melt relative to the axis of rotation of the centrifugal disc. If the angle is too small, the amount of melt splashing will increase because the melt impacts the subsequent wheel too high. If the angle is too large, the melt cannot make sufficient contact with the subsequent wheel, resulting in lower manufacturing efficiency.

[0004] Typically, operators will monitor the manufacturing process and attempt to minimize waste and maximize efficiency. Many adjustments can be made to the fiberizing equipment, such as the location of the trough from which the melt is poured into the centrifuge trays. Attempting to optimize the process by manually making these adjustments requires careful monitoring. Furthermore, determining which adjustments are needed is challenging because the process is difficult to visually monitor due to the size of the fiberizing equipment.

[0005] It is known to use cameras to capture video of the molten flow inside the equipment, allowing operators to view it remotely. However, this still requires skilled image evaluation to determine if adjustments are necessary, and manual intervention by the operator is still required to make those adjustments. Therefore, efficient operation of the equipment cannot be guaranteed.

[0006] Therefore, there is a need for better control over the manufacturing process while reducing the amount of operator input required. This needs to be achieved while maintaining output quality and a safe working environment around the equipment. Summary of the Invention

[0007] According to a first aspect, a method is provided for adjusting at least one fiberization parameter of a fiberization apparatus used in mineral wool manufacturing, the method comprising: generating shape information of a centrifuge disc of the fiberization apparatus by imaging a mark on the fiberization apparatus and mapping the position of the imaged mark to its corresponding physical position using image analysis; quantifying physical melt parameters relative to the centrifuge disc by analyzing the imaged centrifuge disc on which melt to be fiberized is falling based on the shape information and using a computer vision system; and calculating a fiberization parameter and a target value based on the melt parameters, the fiberization parameter depending on the melt parameters, and adjusting the fiberization apparatus to reduce the deviation from the target value if the fiberization parameter deviates from the target value.

[0008] This method allows for better control of the manufacturing process while reducing the required operator input. Calibrating computer vision by generating shape information allows for the calculation of three-dimensional (3D) quantities, such as the impact position of the melt on a centrifugal disc, using only a single imaging device. Calculating melt parameters via computer vision enables continuous process monitoring without requiring a large number of operators or a significant amount of operator time. Furthermore, calculating target values ​​for fiberization parameters and subsequent adjustments to the fiberization equipment further reduces the need for operator input while increasing control over the fiberization process compared to manual operator adjustments.

[0009] In some examples, the marker is located on the fiberizing device, such as on the support structure of the spinner. In other words, in these examples, the marker is not located on the spinning wheel, but rather on a part of the fiberizing device other than the spinning wheel. In these examples, the position of the marker is determined by computer vision, thereby determining the position of the fiberizing device in the field of view of the imaging device. In some examples, the shape information of the spinner is generated with reference to the position of the marker and preset data about the spinner or spinning wheel. In some examples, the preset data may be a 3D model or 3D mapping of the spinner or spinning wheel. In other cases, the preset data may be the position of a previously captured marker. In such examples, the shape information can be generated by mapping the preset data into the field of view of the imaging device using the position of the fiberizing device in the field of view of the imaging device (as determined based on the position of the marker).

[0010] In some examples, the preset data may be generated by an AI model that reconstructs an accurate 3D model of the centrifuge disc and / or rotating wheel using a radiation field (“NERF”) or Gaussian splash. Alternatively, or additionally, the preset data may include measurements of the centrifuge disc and / or rotating wheel using LiDAR, RADAR, or other ranging or range measurements.

[0011] In other examples, the markings are located on the centrifugal disc or the rotating wheel. These examples are discussed further below, but it should be understood that all the following examples are intended to be considered in conjunction with the preceding examples where applicable.

[0012] Imaging devices used to observe melt flow can be calibrated by generating shape information of the centrifuge disc. Shape information typically includes the location of specific points on the centrifuge disc, the location of the disc's edges, and / or a perspective transformation from two-dimensional (2D) points in the camera's field of view to 3D points in space. Typically, markings on the centrifuge disc include calibration patterns on its surface. These calibration patterns can be integral parts of the centrifuge disc itself or placed on the surface. Whether integral or placed on the disc surface, the calibration pattern can mark at least one (or said) physical edge of the centrifuge disc or its vertical diameter axis. During shape information generation, these edges and / or vertical diameter axes can be detected, allowing distances, such as distances from the disc's edges, to be calculated. Calculating these distances provides a mechanism for ensuring the melt flow remains between the edges of the centrifuge disc. Additional markings or geometric patterns may be present, allowing the shape, size, and position of the centrifuge disc to be evaluated and considered in computer vision algorithms.

[0013] Calibration patterns can be regular patterns, such as checkerboard patterns, or patterns formed by arrays or grids of repeating elements, icons, or images. Alternatively, patterns can be irregular patterns that do not repeat, such as gradients, irregular arrays of elements (which may be the same or different), or more complex images. Calibration patterns can include repeating and non-repeating elements, for example, formed by combinations of any of the above elements, respectively.

[0014] The calibration pattern can be applied to the surface of the centrifuge disc as a physical coating, covering, wrapping, or layer. Alternatively, the calibration pattern can be formed by projecting (light) onto the centrifuge disc. In this case, the calibration pattern can be formed on the surface of the centrifuge disc using a projector or other lighting techniques. In one example, a laser projector is used to form the calibration pattern on the surface of the centrifuge disc. In this case, the calibration pattern can be a structured light grid projected by a laser; however, of course, any other calibration pattern discussed above can also be projected onto the surface of the centrifuge disc. The laser projector can be located next to the camera or in a different location.

[0015] Shape information can be generated from a single image of the centrifuge disc, but more commonly, it can be generated from multiple images of the centrifuge disc. Shape information can be generated by any implementation on a computer, including through classical algorithms, machine learning, or AI algorithms. In some examples, AI models using radiation fields (“NERF”) or Gaussian splashes can be used to reconstruct accurate 3D models of the centrifuge disc and / or rotating wheel.

[0016] Computer vision can be used to calculate many melt parameters, such as melt flow stability or the presence of splash. Typically, calculations include at least one physical melt parameter, such as the point of impact of the melt on the centrifugal disc. Monitoring the impact point of the melt is beneficial because the quality of the cotton and the amount of waste generated vary considerably with the impact point.

[0017] Additionally or alternatively, the position of the axial edge of the melt flow relative to the axis of rotation of the centrifugal disk can be calculated, for example, by means of at least one physical melt parameter that includes the axial edge of the melt flow relative to the axis of rotation of the centrifugal disk. Measuring the axial edge position allows detection of the melt flow becoming misaligned with the centrifugal disk and flowing over or across the edge of the centrifugal disk instead of flowing directly onto the surface. Detecting misalignment is beneficial because misaligned melt flow is undesirable and leads to waste.

[0018] Using the measured melt parameters, target values ​​for the fiberization parameters of the fiberization equipment are calculated. These target values ​​are typically calculated to reduce waste and improve output quality. One of the at least one fiberization parameter that can be monitored and may have a target value is the distance between the axial edge of the melt flow (i.e., at least one axial edge) and the axial edge of the centrifugal disc. In this case, the target value can be set to maintain the melt flow at a specified distance from the disc edge. Another example is the location of the melt channel (i.e., the channel from which the melt flows onto the centrifugal disc). Therefore, one of the at least one fiberization parameter may include the location of the channel from which the melt flows onto the centrifugal disc. This target value can be calculated as a function of the melt's impact position, with the aim of maintaining the impact position optimally positioned to minimize waste. Therefore, the target value for at least one fiberization parameter can also be the melt's impact position on the centrifugal disc. Calculating these target values ​​allows for more precise equipment adjustment compared to visual assessment.

[0019] The optimal impact position of the melt can be found by considering the angle formed between the vertical diameter axis of the centrifugal disc and the impact position of the melt relative to the axis of rotation of the centrifugal disc. Typically, when at least one of the fibrillation parameters includes the location of the channel (from which the melt flows onto the centrifugal disc) and the markings on the centrifugal disc include calibration patterns on the surface of the centrifugal disc (marking at least one physical edge of the surface), the target value of the at least one fibrillation parameter, when it is the impact position of the melt on the centrifugal disc, is an impact position where the angle formed between the vertical diameter axis of the centrifugal disc and the impact position of the melt relative to the axis of rotation of the centrifugal disc is in the range of 15 to 35 degrees (°), and the angle is in the same direction as the rotation of the centrifugal disc. The advantage of calculating an angle rather than an absolute position is that it is independent of the size of the centrifugal disc.

[0020] Other fiberization parameters that can be calculated include the energy efficiency of the fiberization equipment (the target value of which can be at least a predetermined amount of efficiency) or the surface damage level of the centrifuge disc itself (where the target can be set to ensure uniform wear of the centrifuge disc across the entire surface).

[0021] The calculation of fiberization parameters and their corresponding target values ​​can be based directly on measured melt parameters, but is typically alternatively based on the average of these melt parameters over multiple measurements. Therefore, at least one fiberization parameter can be calculated based on the moving average of the physical melt parameters. Since the movement of the melt flow can vary considerably, taking the average of melt parameters (such as their position or width) improves the reliability of the fiberization process and results in fewer unnecessary or erroneous adjustments to the equipment.

[0022] If the fiberization parameters deviate from the corresponding calculated target values, the fiberization equipment can be adjusted to reduce the discrepancy. In the case of the melt bath position, a target position can be calculated to maintain the melt impact position optimally. Adjusting the fiberization equipment then involves appropriately moving the melt bath, typically via a hydraulic system or a stepper motor. Therefore, adjusting the fiberization equipment can include adjusting the bath position using a hydraulic system or a stepper motor. The mechanical complexity required to move the melt bath is lower than that required to move a centrifugal disc. Automated movement of the melt bath reduces the amount of operator input required and allows for more frequent adjustments, faster compensation for melt flow variations, and a more efficient manufacturing process.

[0023] In the case of imaging the melt flow via a camera, the camera is typically positioned in a fixed location, providing a constant field of view encompassing the centrifugal disc and the melt. The shape information used to calibrate the computer vision system depends on the camera's position, thus it is desirable to know if the camera has moved in order to recalculate the shape information or invalidate it. Detecting camera movement can be achieved by including a calibration target (which may be attached to the fiberization equipment) within the camera's field of view and using image analysis to identify the calibration target's position within the field of view. Therefore, the imaging is performed via a camera in a fixed position, the camera's field of view includes the calibration target, and the method further includes: using image analysis to identify the position of the calibration target within the field of view, and detecting camera movement based on the image analysis. If the position of the calibration target changes, it can be inferred that the camera has moved. Typically, a human operator will be alerted if the camera is misaligned. This improves the reliability of the manufacturing process and enhances the safety of human operators around the equipment.

[0024] In some examples, adjustments to the fiberizing equipment can be performed automatically, either mechanically, electrically, electromechanically, hydraulically, or pneumatically. Alternatively, in some cases, adjustments to the fiberizing equipment can be performed by a human operator. Therefore, to assist the operator, in some examples, the method may also include outputting calculated fiberizing parameters and the target value, based on which the fiberizing equipment can be adjusted to reduce deviations from the target value.

[0025] The output can be directly to a screen, monitor, or any display device. Alternatively, the output can be delivered in other ways, including via network transmission, physical readings (such as a dial), or via light or sound. Typically, the output is delivered via a screen, which may also display camera output as well as labels representing fiberization parameters and target values.

[0026] To improve the accuracy of shape information, in some examples, generating shape information involves imaging a second marker on the fiberization device and using image analysis to map the position of the imaged second marker to its corresponding physical location. This can improve the accuracy of shape information by helping to locate the centrifuge disc in space.

[0027] In some examples, the melt flows from the tank onto the centrifuge disc. In such examples, it is common for (at least a portion) of the melt to cool as it flows out of the tank, forming "slag"—cooled melt that remains adhered to the tank. This can interfere with the performance of the fiberizing equipment, restricting or disrupting the flow of the melt. Therefore, in some examples, a fiberizing parameter (possibly one of several) includes the amount of slag formed on the tank from which the melt flows onto the centrifuge disc.

[0028] Furthermore, in some examples, the steps of adjusting the fiberizing equipment include removing slag from the tank from which the melt flows onto the centrifugal disc. Slag can be removed mechanically, electromechanically, hydraulically, or pneumatically, or it can be removed manually. In some examples, a robotic arm is used to remove slag by passing the arm through the melt flow and striking the slag. In this way, the arm is used to dislodge the slag from the tank, thereby removing it.

[0029] According to a second aspect, a system is provided for adjusting at least one fiberization parameter of a fiberization apparatus used in mineral wool manufacturing, the system comprising: an imager arranged to image the fiberization apparatus; a calibration mark on the fiberization apparatus; and a processor. The processor is configured to: generate shape information of a centrifuge disc of the fiberization apparatus by imaging the mark on the fiberization apparatus and mapping the position of the imaged mark to its corresponding physical position using image analysis; quantify physical melt parameters relative to the centrifuge disc by analyzing the imaged centrifuge disc on which melt to be fiberized falls based on the shape information and using a computer vision system; and calculate a fiberization parameter and a target value based on the melt parameters, the fiberization parameter depending on the melt parameters, and adjust the fiberization apparatus to reduce the deviation from the target value if the fiberization parameter deviates from the target value. Attached Figure Description

[0030] Exemplary processes and exemplary devices are described in detail herein with reference to the accompanying drawings, in which:

[0031] Figure 1 A schematic diagram of an exemplary device is shown;

[0032] Figure 2a , Figure 2b and Figure 2c A schematic diagram of the components required for calibration is shown;

[0033] Figure 3 A flowchart of the method is shown;

[0034] Figure 4 shows an example of training data used to train a computer vision algorithm to determine melt parameters;

[0035] Figure 5 Another example of training data used to train a computer vision system to generate shape information is shown;

[0036] Figure 6 An example of a computer vision system detecting the width of a melt flow is shown;

[0037] Figure 7 Four examples of training data are shown for training a computer vision system to detect the physical boundaries of a melt flow impacting a rotating wheel;

[0038] Figure 8 Six additional examples of labeled training data are shown, illustrating specific rotating wheels that can be used to train computer vision systems to detect the location and physical boundaries of melts;

[0039] Figure 9 It shows the source and Figure 8Six more examples of labeled training data with the same camera and rotating wheel;

[0040] Figure 10 Six additional examples of labeled training data are shown, illustrating specific rotating wheels that can be used to train computer vision systems to detect the location and physical boundaries of melts;

[0041] Figure 11 It shows the source and Figure 10 Six more examples of labeled training data with the same camera and rotating wheel;

[0042] Figure 12 Six additional examples of labeled training data are shown, illustrating specific rotating wheels that can be used to train a computer vision system to detect the location and physical boundaries of melt flows; and

[0043] Figure 13 It shows the source and Figure 12 Six more examples of labeled training data using the same camera and rotating wheel. Detailed Implementation

[0044] An exemplary system including a fiberizing device and a system for adjusting at least one fiberizing parameter of the fiberizing device is described in Figure 1 The system is generally shown as 1. It has two centrifuge discs 10 and 11 (also referred to as the "first centrifuge disc" and the "second centrifuge disc", respectively), on which melt 30 is being poured from melt tank 20.

[0045] The melt is generated in a furnace and, in some arrangements, gradually cools through a series of tanks before reaching the melt bath. Centrifugal discs typically rotate at high speeds. In various examples, this rotational rate is approximately 7000 RPM (revolutions per minute). This causes the melt to become fiberized, producing cotton.

[0046] The falling melt 30, the first centrifuge disc 10, and the calibration target 51 are all within the field of view 41 of the camera 40. The camera can be used to capture images as video or (freeze-frame) image feeds. The images captured by the camera are transmitted to the processor 50. Figure 1 In the example shown, image transmission uses a wired connection. In other examples, transmission may include or may be performed entirely wirelessly.

[0047] The system is capable of measuring one or more melt parameters, determining fiberization parameters, calculating target values ​​for these fiberization parameters, and then adjusting the fiberization equipment if the fiberization parameters differ from the target values ​​by using the processor 50 to perform computer vision analysis on the images captured by the camera 40.

[0048] In various examples, adjustments to the fiberizing equipment involve changing the position of the melt tank 20. Adjusting the position of the tank changes the impact point of the melt 30 on the centrifugal disc 10. The impact point of the melt is important for optimizing the manufacturing process. This is because optimally positioning the melt impact point reduces waste and equipment wear. Therefore, in some examples, one of the fiberizing parameters is taken as the melt impact angle 25 in the vertical direction 13 (or perpendicular direction) relative to the axis of rotation of the centrifugal disc, and the target value is set at an optimal value, within the range of 15 to 35 degrees.

[0049] In order to allow the melt tank 20 to move, in some examples the melt tank is connected to a moving mechanism 23, which allows it to move parallel to the axis of rotation of the centrifugal disc 10 and parallel to the axis defined by the direction of rotation of the top of the centrifugal disc, while maintaining a constant vertical distance from the disc.

[0050] In one example, the moving mechanism 23 may include a set of tracks. In one example, the set of tracks may include tracks that allow the melt tank to move parallel to the axis of rotation of the centrifugal disc. In another example, the set of tracks may include tracks that allow the melt tank to move parallel to an axis defined by the direction of rotation of the top of the centrifugal disc. In yet another example, the tank may also move freely along a vertical axis separating it from the centrifugal disc.

[0051] The moving mechanism 23 is connected to the controller 21, which moves the slot in response to a signal from the processor 50. Figure 1 In this example, the signal is transmitted using a wired connection. In other examples, signal transmission may include a wireless connection or may be entirely wireless.

[0052] In some examples, the controller can move the slot via an actuator (such as a linear actuator) or a hydraulic system. In other examples, the movement of the slot is achieved using a stepper motor or other means.

[0053] In some examples, to detect whether camera 40 has maintained the same position and orientation since the shape information was generated, the position of target 51 within the camera's field of view 41 is recorded. In various examples, computer vision is used to detect the target's position each time a measurement is performed or each time the camera is operated. If the target's position has changed, it is inferred that the camera's position or orientation has changed, and therefore the generated shape information is no longer valid. If the shape information is no longer valid, the measurements performed by the system cannot be considered accurate. In some examples, this means that adjustments to the fiberization equipment are disabled, and the operator is notified.

[0054] Before any melt parameter measurements can be performed, system 1 must be calibrated by generating shape information. Figure 2a , Figure 2b and Figure 2cAt least a portion of the view of the camera 40 in the field of view 41 is depicted as a series of synthetic images, showing information related to the generation of shape information.

[0055] exist Figure 2a The image depicts a centrifugal disc 10, from which melt 30 flows out of a melt tank 20, as seen from camera 40. Figure 2c The same field of view is shown in the image, but without the melt being poured onto the centrifuge dish. Instead, a centrifuge dish is shown with a calibration pattern 60 applied to its outer surface. Figure 2c In the example shown, the pattern is a black and white checkerboard pattern. In other examples, the pattern may be integrated with the centrifugal disc rather than applied separately. Additionally or alternatively, in other examples, the pattern may also be a pattern different from the checkerboard pattern, such as a test pattern or a resolution pattern.

[0056] The pattern is detected by a computer vision algorithm running on processor 50 and used to generate shape information describing the mapping between 2D pixels in the camera's field of view and points in 3D space. In some examples, an applied checkerboard pattern is used. Corner detection algorithms (such as Moravec, Harris, or SUSAN corner detection algorithms) are used to detect the corners of the squares in the checkerboard pattern. In other examples, binary-based or contour-based corner detection methods can be used. Still in some examples, a machine learning algorithm or neural network can be trained based on the labeled data of the checkerboard pattern and used to detect the corners of the checkerboard pattern. In other examples, the calibration pattern may not include the checkerboard pattern, and other properties of the pattern, such as edges or specific features, can be detected.

[0057] After detecting the positions of the corner points of the checkerboard pattern, the position of each corner point in the camera's 2D field of view can be mapped to its known position in 3D space. In one example, this mapping includes calculating a rotation matrix and translation that maps the camera coordinate system with its origin at the camera's optical center to a 3D coordinate system with its origin at the center of the centrifugal disk. Additionally or alternatively, this mapping includes calculating a skew coefficient to compensate for the camera's non-perpendicular image axis. Additionally or alternatively, lens distortion of the camera is compensated based on the lens's focal length. In some examples, this distortion can be radial or tangential distortion.

[0058] exist Figure 2b The image shows a composite image of a centrifuge disc 10 with a calibration pattern 60 applied and a centrifuge disc with falling melt 30. A computer vision system running on a processor 50 is used to detect the corner points of the checkerboard pattern and locate the edges 55 of the checkerboard pattern from the corner points.

[0059] In various examples, the width 53 of the falling melt is also determined. The position of the falling melt 30 in the image captured by camera 40 is determined by machine learning, which will be described in detail later. The position of the falling melt in the image is transformed into 3D space using shape information. The width is given by the difference between the positions of the two edges of the melt impacting the centrifuge disc 10, based on the known width of the physical centrifuge disc (i.e., the distance from front to back). In some examples, the width of the melt can be determined at the point of impact.

[0060] In addition to the checkerboard calibration pattern, Figure 2c In the example shown, the applied calibration pattern 60 also includes edge markers 61 and top markers 63. Edge markers are lines extending around the circumference of the centrifuge dish and indicating the location of the outer edge of the centrifuge dish. In another example, system 1 can be configured to alternatively detect markers along the inner edge of the centrifuge dish.

[0061] Edge marker 61 is detected by a computer vision algorithm running on processor 50. This detection is used to calculate the distance from the edge of the disc 10.

[0062] Top mark 63 is a line along the axis of rotation of the centrifuge disc, indicating the very top of the disc. The top mark is applied manually to the first centrifuge disc 10 or as part of the disc manufacturing process. The top mark is applied as part of, supplement to, or independently of the calibration pattern 60.

[0063] During calibration setup, the top mark 63 is positioned at the top (i.e., the highest point or surface) of the first centrifuge disc 10. In some examples, this is achieved by matching or aligning with a mark (not shown) on the fiberization device, but in other examples, it can be simply aligned visually (i.e., without using a mark or label). The top mark is detected by a computer vision algorithm and used to help define the extent of the centrifuge disc.

[0064] Precise positioning of the markers is important for system performance. In one example, placing the edge marker 61 with an accuracy of ±2 mm and aligning the top marker with a radial accuracy of ±2° yielded sufficiently good performance.

[0065] Calibration data was obtained using exposure settings on the camera that differed from those used when imaging the melt flow. This is because the melt is significantly brighter than the centrifuge disc under ambient conditions due to its temperature, thus requiring higher sensitivity during calibration measurements.

[0066] exist Figure 3 An exemplary schematic is shown in the figure. Figure 3The document details the steps of an exemplary process for adjusting fiberization parameters. The process begins with a calibration step 70, in which shape information is generated by imaging a centrifuge disc 10 with calibration marks 60 and processing the image with a computer vision system running on a processor 50, as described above regarding... Figures 2a to 2c As stated above.

[0067] Once the system is calibrated, repeat in sequence. Figure 3 The remaining steps are shown until the process terminates. These steps include image capture step 71, optional camera misalignment step 72, melt parameter step 73, fiberization parameter step 75, and adjustment step 77.

[0068] Image capture step 71 includes capturing an image with camera 40 and transmitting the image to processor 50. The image can then be analyzed using machine learning algorithms in a computer vision system, as will be explained in detail later.

[0069] The camera misalignment step 72 includes measuring the position of the calibration target 51 in the field of view 41 of the camera 40 using a computer vision system running on the processor 50. During positioning, it is identified whether the position of the calibration target in the field of view is within a predetermined threshold of the position measured during calibration step 70.

[0070] Using a small threshold allows for some minor shifts in the camera position due to vibrations of device 1, but does not tolerate large deviations. Therefore, it is confirmed that camera 40 has not moved since the shape information was generated.

[0071] If camera 40 is detected to have moved, the process is terminated in some examples. This is because shape information can no longer be relied upon to convert between the 2D image and the 3D quantity being measured, leading to inaccurate measurements. This inaccuracy can result in incorrect adjustments to the system, potentially reducing efficiency or jeopardizing equipment or the operator.

[0072] In the melt parameter step 73, the physical parameters of the melt are determined by a computer vision system running on processor 50 and trained with appropriate training data. In this example, the impact position of melt 30 on centrifugal disk 10, as well as the width of the melt and the distance of the melt from the edge of the disk along the axis of rotation are calculated.

[0073] In some examples, the measurement of the impact position is accurate enough to determine the melt impact angle 25° to ±2°. Here, the width of the melt is measured at the impact position of the melt on the centrifuge disc. The melt impact angle is determined relative to the center of the measured width line 53. A disadvantage of this is that the camera's field of view may be obstructed by slag present in the flow. Therefore, in other examples, additionally or alternatively, the width of the melt can be measured below the impact position, i.e., at an increased angle relative to the impact angle 25°. Measurements taken below the impact position are less likely to be obstructed by slag, as the slag must be larger. However, this can be inaccurate due to geometric distortion as the melt detaches from the surface of the subsequent centrifuge disc 11. Shape information cannot account for this detachment, therefore the width measurement may be incorrect.

[0074] In the fiberization parameter step 75, on the processor 50, the fiberization parameters are calculated as a function of the melt parameters found in the melt parameter step 73. In this example, one of the fiberization parameters is the melt impact angle 25, which is the angle formed between the vertical diameter axis 13 of the centrifugal disk 10 and the impact position of the melt relative to the rotation axis of the centrifugal disk 10, and the angle is in the same direction as the rotation direction of the centrifugal disk. The second fiberization parameter is the distance between the melt and the radial edge of the centrifugal disk, the position of which is obtained from the shape information generated by the self-calibration mark 61.

[0075] Each of these fiberization parameters has an associated target value, which is then calculated on processor 50. In other examples, there may be fiberization parameters without a target value. The target value may be known in advance or may need to be calculated based on external factors, depending on the desired outcome.

[0076] In some examples, target values ​​for the fiberization parameters can be calculated to maximize energy efficiency or minimize wear on the equipment during long-term operation. In this example, the target value for the fiberization parameters is set to minimize waste and optimally position the melt flow. Therefore, a target value for the melt impact angle 25° is determined to minimize the amount of waste generated, approximately between 15 and 35 degrees.

[0077] In another example, target values ​​for the fiberization parameters can be calculated to ensure that the melt does not flow past the front or rear of the centrifuge disc 10 (i.e., the edge relative to the axis of rotation). The target distance between the melt and the edge of the centrifuge disc is set to ensure sufficient distance between the melt and the edge of the centrifuge disc.

[0078] In other examples, the target value can be set by the operator, causing the system to automatically maintain the user-specified fiberization parameter value. This means that the fiberization parameters of the fiberization equipment can be manually controlled, but require less input from the operator compared to manually adjusting to ensure that the fiberization parameters remain constant.

[0079] In adjustment step 77, the fiberizing equipment is adjusted if necessary. In some examples, to determine whether to adjust the equipment, the processor 50 compares the difference between the fiberizing parameters and the target value of the fiberizing parameters with a threshold.

[0080] In various examples, the target value is compared to a moving average of the fiberization parameters over a small number of measurements. Using the moving average reduces the impact of erroneous measurements and compensates for random shifts in the melt flow. This results in a delay in several examples, such as approximately 10 seconds, before the fiberization equipment can be adjusted. This delay in adjustment can mean lower system efficiency, as misaligned melt flow will be tolerated for a short period before adjustment is made.

[0081] Different fiberization parameters can have different thresholds. Different thresholds help the process consider whether the fiberization equipment is more or less sensitive to a particular fiberization parameter.

[0082] If the difference between a specific fiberization parameter and its corresponding target value exceeds a threshold, the equipment is adjusted. In this example, the required adjustment is the movement of the melt bath 20. The processor sends a signal to the controller 21, which mechanically moves the melt bath along axis 23 and repositions the melt flow.

[0083] In this example, the melt tank 20 is moved using hydraulic technology. In various examples, the system is equipped with sensing sensors to ensure smooth operation and prevent any deviation beyond the optimal range of the hydraulic system.

[0084] In some examples, the process is configured to make small, incremental changes to assess the impact of the changes on the system and avoid overcompensation and missing the target value. However, in other examples, such as with electronic parameters (e.g., voltage), the device can be adjusted to the desired value immediately.

[0085] Making small changes and repeatedly reassessing the condition of the fiberizing equipment means that only the relative values ​​of the fiberizing equipment adjustments are needed, not the absolute values.

[0086] In some examples, the computer vision system uses machine learning algorithms to identify melt parameters from images captured by camera 40. These melt parameters include the impact position of the melt on centrifuge disc 10 and the position of the edge of the falling melt.

[0087] exist Figure 4a , Figure 4b and Figure 4cThe image shows an example of training data used to train a machine learning algorithm in a computer vision system. Machine learning algorithms are image classifiers capable of recognizing elements in an image with a given degree of certainty (e.g., as a function of how well an element matches an element in the training data, expressed as a percentage). These algorithms typically require a large corpus of labeled training data, which provides image pairs and the expected output from the algorithm for that image, and a statistical mapping between the training data images and the expected output. Once this training process is complete, the machine learning algorithm is trained and no longer requires access to the corpus of training data.

[0088] exist Figure 4a The image shown is captured by camera 40, depicting melt 30 falling from melt tank 20 onto first centrifuge disc 10. Also shown as a line (red in the original submission figures) is label 81, which marks the impact position of the melt across the full width of the melt on the centrifuge disc. A large series of such labeled images (i.e., the expected output from the machine learning component of the computer vision system) form a corpus of training data for detecting the impact position of the melt. Figure 4b The image shows a second example of this type, with a slightly more complex geometry, as the bright splash 86 is excluded from the labeled area.

[0089] Training data that includes anomalies (such as splashes and other materials) helps the algorithm adapt to their presence in the image. As another example, Figure 4c The diagram illustrates a situation where a large amount of slag 83 (an example of other materials that can be considered anomalous) obstructs the camera's field of view of the melt flow. In this case, the marked area 81 is divided into several parts, and the edges of the melt flow are more difficult to detect.

[0090] exist Figure 5 The diagram shows training data used to generate shape information. In the upper panel, a rotating wheel 10 is shown, its surface having a calibration pattern 60. In this example, a checkerboard calibration pattern has been applied to the rotating wheel. It provides edge markers 61 (shown in red in the original submission's drawing) and top markers 63 (shown in red in the original submission's drawing). In this example, as shown in the lower panel, the computer vision system detects the position of the top marker 63. This is shown by green line 57 in the original submission's drawing. The computer vision system further detects the positions of the corner points of the checkerboard pattern, shown by red dots 54 in the original submission's drawing. This allows the shape information to be calculated as described above.

[0091] exist Figure 6The diagram illustrates the detection of the impact position of melt 30 on centrifugal disc 10 in the presence of slag in the melt flow. The shaded area 87, shown in orange, indicates the region of falling melt that the computer vision system determined had not yet impacted the centrifugal disc. The shaded area 89 (shown in blue in the original submission) indicates the region of melt that the computer vision system determined had impacted the centrifugal disc. The computer vision system was able to determine this in part due to the shape information used to calibrate the system, the key point of which is shown as line 55 (shown in orange in the original submission). The computer vision system had estimated the width 53 of the melt flow, which, due to the interference of slag in the measurement, did not span the full width of the melt.

[0092] Figure 7 Four additional examples of labeled training data are shown, which can be used to train a computer vision system to detect the location and physical boundaries of melt flows. In these examples, the images are labeled by marking two regions on the original image. A falling melt region 87 marks the area where the melt has not yet impacted the rotating wheel. A spun melt region 89 marks the area where the melt has impacted the rotating wheel. Many such images are needed to train the computer vision system to recognize these regions on arbitrary input. By using training data from a large number of different cameras and rotating wheels, the ability of the computer vision system to classify rotating wheels without prior training can be developed.

[0093] Figure 8 Six additional examples of labeled training data are shown, which can be used to train a computer vision system to detect the location and physical boundaries of melt flows. In these examples, images are labeled by marking regions on the original images. Falling melt region 87 marks the area where the melt has not yet impacted the rotating wheel. Centrifuged melt region 89 marks the area where the melt has impacted the rotating wheel. Six additional examples from the same rotating wheel and camera can be used... Figure 9 I saw it in the middle.

[0094] Figure 10 Six additional examples of labeled training data are shown, which can be used to train a computer vision system to detect the location and physical boundaries of melt flows. Six more examples from the same rotating wheel and camera can be used... Figure 11 I saw it in the middle.

[0095] Figure 12 Six additional examples of labeled training data are shown, which can be used to train a computer vision system to detect the location and physical boundaries of melt flows. Six more examples from the same rotating wheel and camera can be used... Figure 13 I saw it in the middle.

[0096] Figures 7 to 12Each of the labeled training datasets shown is a subset of a larger labeled training dataset. The raw images of the labeled training dataset are collected and labeled to identify areas of falling melt and melt that have reached the centrifuge pan. These are then fed to a computer vision system for training.

[0097] In total, the labeled training dataset includes approximately 2000 images. Figures 7 to 12 As can be seen, many different scenes were incorporated into the labeled training data. These include "clean" images, such as... Figure 7 The top row and bottom left corner images are shown. Other images in the labeled training data include splashes, for example... Figure 7 The image shown is in the lower right corner. Some images include "shots," which are typically single droplets of melt that have separated from the rest of the melt. See Figure 4 and... Figure 6 As shown in some images, slag is depicted. It forms on the chute.

[0098] Images for the labeled training dataset were collected using a variety of different camera and centrifuge plate relative positions. Throughout the labeled training dataset, approximately ten different camera and centrifuge plate relative positions were used. In various examples, the labeled training dataset included varying numbers of images and / or varying numbers of camera-centrifuge plate relative positions.

Claims

1. A method for adjusting at least one fiberization parameter of a fiberization device used in mineral wool manufacturing, the method comprising: The shape information of the centrifuge disc of the fiberization device is generated by imaging the marks on the fiberization device and using image analysis to map the position of the imaged marks to their corresponding physical positions. The physical melt parameters relative to the centrifuge disk are quantified by analyzing the imaged centrifuge disk on which the melt to be fiberized falls, based on the shape information and using a computer vision system. as well as The fiberization parameters and their target values ​​are calculated based on the melt parameters. The fiberization parameters depend on the melt parameters, and if the fiberization parameters deviate from the target values, the fiberization equipment is adjusted to reduce the deviation from the target values.

2. The method according to claim 1, wherein, The markings include those on the centrifuge disc.

3. The method according to claim 1 or 2, wherein, At least one of the physical melt parameters includes the impact position of the melt on the centrifugal disc.

4. The method according to any one of the preceding claims, wherein, At least one of the physical melt parameters includes the axial edge of the melt flow relative to the axis of rotation of the centrifugal disk.

5. The method according to any one of the preceding claims, wherein, One of the at least one fiberization parameters includes the location of the tank from which the melt flows onto the centrifugal disc.

6. The method according to claim 5, wherein, Adjusting the fiberizing equipment includes adjusting the position of the trough using a hydraulic system or a stepper motor.

7. The method according to any one of the preceding claims, wherein, The markings on the centrifuge disc include a calibration pattern on the surface of the centrifuge disc, the calibration pattern indicating at least one physical edge of the surface.

8. The method according to claims 5 and 7, wherein, The target value of the at least one fiberization parameter is the impact position of the melt on the centrifugal disc.

9. The method according to claim 8, wherein, The target value of the at least one fiberization parameter is the impact position, which is the angle between the vertical diameter axis of the centrifugal disc and the impact position of the melt, relative to the axis of rotation of the centrifugal disc, in the range of 15 degrees to 35 degrees, and the angle is in the same direction as the rotation direction of the centrifugal disc.

10. The method according to claims 4 and 7, wherein, The target value of the at least one fiberization parameter is the distance between at least one axial edge of the melt flow and at least one axial edge of the centrifugal disc.

11. The method according to any one of the preceding claims, wherein, One of the at least one fiberization parameter includes the energy efficiency of the fiberization equipment.

12. The method according to any one of the preceding claims, wherein, The imaging is performed by a camera located at a fixed position, the camera's field of view including the calibration target, and the method further includes: using image analysis to identify the position of the calibration target in the field of view, and detecting positional movement of the camera based on the image analysis.

13. The method according to any one of the preceding claims, wherein, The at least one fiberization parameter is calculated based on the moving average of the physical melt parameters.

14. The method according to any one of the preceding claims further comprises: The calculated fiberization parameters and the target value are output, and the fiberization equipment can be adjusted based on the output to reduce the deviation from the target value.

15. The method according to any one of the preceding claims, wherein, The markings are formed by projecting them onto the centrifuge disc.

16. The method according to any one of the preceding claims, wherein, The at least one fiberization parameter includes the amount of slag formed on the tank, from which the melt flows onto the centrifugal disc.

17. The method according to claim 16, wherein, The steps of adjusting the fiberizing equipment include: removing slag from the tank and allowing the melt to flow from the tank onto the centrifugal disc.

18. A system for adjusting at least one fiberization parameter of a fiberization apparatus used in mineral wool manufacturing, the system comprising: An imager is arranged to image the fiberization device; Calibration markings on the fiberizing equipment; as well as The processor is configured as follows: The shape information of the centrifuge disc of the fiberization device is generated by imaging the marks on the fiberization device and using image analysis to map the position of the imaged marks to their corresponding physical positions. The physical melt parameters relative to the centrifuge disk are quantified by analyzing the imaged centrifuge disk on which the melt to be fiberized falls, based on the shape information and using a computer vision system. as well as The fiberization parameters and their target values ​​are calculated based on the melt parameters. The fiberization parameters depend on the melt parameters, and if the fiberization parameters deviate from the target values, the fiberization equipment is adjusted to reduce the deviation from the target values.

Images