A method and device for monitoring the traditional Chinese medicine fluidized bed granulation process based on a robot

Abstract

The present invention discloses a method and device for monitoring the Chinese medicine fluidized bed granulation process based on a robot, which relates to the technical field of Chinese medicine production and manufacturing. The method includes detecting the particle moisture and particle size parameters in the granulation process based on a first robot; inspecting the pressure and temperature parameters of the fluidized bed equipment based on a second robot; and collaboratively controlling the first robot and the second robot to implement the monitoring of the granulation process. The device includes a robotic arm, a computer vision mechanism, a traveling mechanism, and an alarm mechanism; the robotic arm and the alarm mechanism are installed at the upper end of the traveling mechanism, and the alarm mechanism is composed of a warning light and a voice prompt mechanism; the computer vision mechanism is installed at the upper end of the robotic arm. The present invention can autonomously inspect the temperature and pressure conditions of the fluidized bed equipment during the granulation process through the robot, timely detect abnormal equipment operation, alarm and suspend the equipment operation, and avoid production safety risks.

Description

Technical Field

[0001] The present invention relates to the technical field of traditional Chinese medicine production and manufacturing, and particularly to a monitoring method and device for the fluidized bed granulation process of traditional Chinese medicine based on a robot. Background Art

[0002] At present, the one-step fluidized bed granulation method has been widely used in the production of traditional Chinese medicine. This method integrates mixing, granulation, and drying in a closed container. Compared with other granulation methods, the granules prepared by the fluidized bed granulation method have better fluidity, uniformity, and compression molding properties. However, due to the complex composition of traditional Chinese patent medicine, the current fluidized bed granulation process is still unstable, and there are problems such as difficulty in monitoring the quality attributes of granules and difficulty in timely detecting safety risks in the granulation process.

[0003] The moisture content and particle size of granules are two key quality attributes of the fluidized bed granulation process, which are closely related to subsequent production processes and the content of active pharmaceutical ingredients. In addition, the fluidized bed granulation process requires heating and pressurization. Therefore, on-site real-time monitoring of various pressure parameters and equipment temperatures during the granulation process is particularly important for safe production. At present, in the process of traditional Chinese medicine fluidized bed granulation, workers need to manually take out the granules and use traditional particle size meters and moisture meters to measure the particle size and moisture respectively. This method is cumbersome and time-consuming, and cannot meet the requirements of rapid detection in the pharmaceutical production process. At the same time, the monitoring of the temperature and pressure of the fluidized bed equipment also requires inspections and operators to record and adjust, which poses certain safety hazards. In order to quickly detect the moisture and particle size of granules during the fluidized bed granulation process and avoid safety risks to a greater extent, a monitoring method for the fluidized bed granulation process that can perform self-execution, self-sensing, and self-decision-making is needed.

[0004] Chinese Patent No. CN106546516A discloses an on-line detection device for multiple properties of granules during the fluidized bed granulation process. The detection device is installed outside the fluidized bed reaction chamber. The detection device has a sealed housing, a sample collection mechanism capable of taking out samples from the reaction chamber, and a vision detection unit, a moisture detection unit, and a density detection unit located inside the housing for photographing the falling process of the taken-out samples inside the housing and detecting the density of the granules; the vision detection unit includes a camera and a graduated cylinder; the graduated cylinder receives the samples falling from the sample collection mechanism, and the camera photographs the process of the samples falling into the graduated cylinder and the image of the sample accumulation formed on the top of the graduated cylinder when the sample overflows the graduated cylinder. The image processing system obtains the particle size information from the falling image of the sample and the angle of repose from the sample accumulation image. However, this patent cannot effectively detect multiple quality indicators during the granulation process and cannot achieve autonomous operation and adjustment. Summary of the Invention

[0005] In view of the problems existing in the prior art, the present invention provides a monitoring method and device for the fluidized bed granulation process of traditional Chinese medicine based on a robot.

[0006] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0007] A method for monitoring the traditional Chinese medicine fluidized bed granulation process based on a robot, comprising the following steps:

[0008] Step S1: Detect the particle moisture and particle size parameters during the granulation process based on a first robot;

[0009] Step S2: Patrol and inspect the pressure and temperature parameters of the fluidized bed equipment based on a second robot;

[0010] Step S3: Cooperatively control the first robot and the second robot to implement the monitoring of the granulation process.

[0011] Based on the above technical solution, further, the detection process in step S1 includes the following steps:

[0012] Step S11: Collect sample particles;

[0013] Step S12: Collect the hyperspectral images of the sample particles;

[0014] Step S13: Determine the mass parameters of the sample particles;

[0015] Step S14: Establish a quantitative calibration model;

[0016] Step S15: Complete the detection of the fluidized bed granulation process.

[0017] Based on the above technical solution, further, the process of establishing the quantitative calibration model includes the following steps:

[0018] Step S141: Collect sample particles during the fluidized bed granulation process;

[0019] Step S142: Collect the hyperspectral images of the sample particles and calculate the average spectrum;

[0020] Step S143: Determine the particle size and moisture of the sample particles;

[0021] Step S144: Establish a quantitative calibration model for particle moisture, particle size and the content of pharmaceutically active substances by using partial least squares method;

[0022] Based on the above technical solution, further, the process of collecting particles in step S11 is as follows: The first robot moves to the vicinity of the fluidized bed under the guidance of the laser detection device, the signal acquisition device is aligned with the sampling port position of the fluidized bed, and the robotic arm is started. The robotic arm drives the clamping device to move to the position where the sample particles are placed, and the clamping device clamps the sample particles, thereby realizing the collection of the sample particles.

[0023] Based on the above technical solution, further, the process of hyperspectral image acquisition in step S12 is as follows: The sample particles clamped by the clamping device are put into the hyperspectral detection device through the sample inlet. The switch is turned on, and the hyperspectral data of the sample particles are obtained through the imaging system and transmitted to the remote server.

[0024] Based on the above technical solution, further, the average spectrum calculation formula in step S142 is as follows: Where, is the average reflectance of the current sample at wavelength band w, n is the number of pixel points of the hyperspectral image of the current sample, is the average spectrum of the current sample, and then the spectral information of different samples is calculated.

[0025] Based on the above technical solution, further, the inspection method in step S2 includes the following steps:

[0026] Step S21: Equipment image acquisition;

[0027] Step S22: Obtain the temperature of each area of the equipment based on the infrared thermal imaging picture;

[0028] Step S23: Identify the reading of the target pressure gauge;

[0029] Step S24: Determine whether the temperature and pressure of the fluidized bed equipment are normal.

[0030] Based on the above technical solution, further, the process of identifying the reading of the target pressure gauge in step S23 includes the following steps:

[0031] Step S231: Process all the collected pressure gauge images to make a data set;

[0032] Step S232: Based on the YOLO network model, establish a target detection model for pressure gauge images. Input the processed pressure gauge image data set and train the model for the dial, pointer, and reference in the image to obtain a trained pressure gauge image detection model;

[0033] Step S233: Input the image into the trained pressure gauge image detection model, obtain the coordinate position information of the dial, pointer, and reference, and determine the subordinate relationship of the dial, pointer, and reference according to the coordinate values;

[0034] Step S234: According to the coordinates of the dial, pointer, and reference obtained in step S233, calculate the included angle between the line connecting the center of the dial and the center of the reference and the line connecting the center of the dial and the center of the pointer. The calculation formula is:

[0035]

[0036] where angle ∈ (-π, π), the center of the dial is (x c , y c ), the center of the pointer is (x b , y b ), the reference center is (x a , y a ), and d1, d2 are

[0037] Step S235: Scale conversion. Since the line connecting the reference center and the center of the dial bisects the scale, the conversion formula is: number = 0.5 + 0.02 × angle / 5.4, where number is the reading of the pressure gauge and angle is the angle between the line connecting the center of the dial to the reference center and the line connecting the center of the dial to the center of the pointer;

[0038] Step S236: Scale printing. Print the scale information onto the camera screen.

[0039] Based on the above technical solution, further, the collaborative control process in Step S3 includes the following steps:

[0040] Step S31: Map location sharing;

[0041] Step S32: Detection and inspection anomaly alarm;

[0042] Step S33: Fluidized bed equipment regulation.

[0043] Based on the above technical solution, further, the anomaly alarm process in Step S32 is as follows: If the detection and inspection results in Step S15 and Step S24 are abnormal, the abnormal information will be sent to the alarm mechanism of the second robot via the Modbus protocol through the local area network, and the alarm mechanism will alarm through the warning light language and voice.

[0044] Based on the above technical solution, further, it includes a robotic arm, a computer vision mechanism, a traveling mechanism, and an alarm mechanism; the traveling mechanism is installed with a robotic arm and an alarm mechanism at its upper end; the alarm mechanism consists of a warning light and a voice prompt mechanism; the computer vision mechanism is installed at the upper end of the robotic arm.

[0045] Compared with the prior art, the present invention has the following beneficial effects:

[0046] (1) The present invention can autonomously sample by the robot and detect the particle size and moisture in real time during the fluidized bed granulation process, realize the transparency of the fluidized bed granulation process, timely discover abnormal fluctuations of the material, alarm and autonomously adjust.

[0047] (2) The present invention can autonomously inspect the temperature and pressure of the fluidized bed equipment during the granulation process by the robot, timely discover abnormal operation of the equipment, alarm and suspend the operation of the equipment, and avoid production safety risks.

[0048] (3) The present invention can optimize the operating efficiency of robots and maximize the functions of robots through collaborative control of multiple robots, enabling a set of multi-robot collaborative control systems to be shared by multiple production devices, and minimizing the cost investment in the transformation towards unmanned and intelligent production as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is a flowchart of the monitoring method of the present invention;

[0050] Figure 2 is a graph showing the time-sequential change of the water content during the fluidized bed granulation process of Ganoderma lucidum spore powder particles in Example 1 of the present invention;

[0051] Figure 3 is a graph showing the time-sequential change of the particle size distribution (<10%) during the fluidized bed granulation process of Ganoderma lucidum spore powder particles in Example 1 of the present invention;

[0052] Figure 4 is a graph showing the time-sequential change of the particle size distribution (<50%) during the fluidized bed granulation process of Ganoderma lucidum spore powder particles in Example 1 of the present invention;

[0053] Figure 5 is a graph showing the time-sequential change of the water content during the fluidized bed granulation process of Dendrobium officinale Kimura et Migo particles in Example 2 of the present invention;

[0054] Figure 6 is a graph showing the time-sequential change of the particle size distribution (<3%) during the fluidized bed granulation process of Dendrobium officinale Kimura et Migo particles in Example 2 of the present invention;

[0055] Figure 7 is a graph showing the time-sequential change of the particle size distribution (<6%) during the fluidized bed granulation process of Dendrobium officinale Kimura et Migo particles in Example 2 of the present invention;

[0056] Figure 8 is a graph showing the time-sequential change of the particle size distribution (<10%) during the fluidized bed granulation process of Dendrobium officinale Kimura et Migo particles in Example 2 of the present invention;

[0057] Figure 9 is a graph showing the time-sequential change of the particle size distribution (<25%) during the fluidized bed granulation process of Dendrobium officinale Kimura et Migo particles in Example 2 of the present invention;

[0058] Figure 10 is an image of a pressure gauge to be detected in Example 2 of the present invention;

[0059] Figure 11 is a graph showing the detection result of the pressure gauge image in Example 2 of the present invention;

[0060] Figure 12 is a schematic structural diagram of Example 3 of the present invention;

[0061] Reference numerals: 1. robotic arm; 2. computer vision mechanism; 21. 4K high-definition camera; 22. infrared thermal imaging camera; 3. traveling mechanism; 31. laser detection mechanism; 32. first switch; 4. alarm mechanism; 41. warning light; 42. voice prompt mechanism. Detailed implementation manners

[0062] To make the objectives and technical solutions of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below in conjunction with embodiments.

[0063] As Figure 1 shown, a method for monitoring the Chinese medicine fluidized bed granulation process based on a robot includes the following steps:

[0064] Step S1: Detect the particle moisture and particle size parameters during the granulation process based on a first robot; wherein, the first robot is used to quickly detect the particle moisture and particle size in the fluidized bed granulation process, and may include a robotic arm, a clamping device, a signal acquisition device, a transportation device, and a hyperspectral detection device. The working principle of the first robot is as follows: according to the time program, the first robot clamps a fluidized bed particle sample with the clamping device at regular intervals and sends it to the hyperspectral imaging system for imaging. The hyperspectral data is transmitted to the server for calculation to obtain the content information of the main components.

[0065] Specifically, its detection process includes the following steps:

[0066] Step S11: Collect sample particles;

[0067] Step S12: Acquire the hyperspectral image of the sample particles; wherein, the hyperspectral image acquisition process is as follows: put the sample particles clamped by the clamping device into the hyperspectral detection device through the sample inlet, start the switch, obtain the hyperspectral data of the sample particles through the imaging system, and transmit it to the remote server.

[0068] Step S13: Determine the mass parameters of the sample particles; the mass parameters at least include particle size and moisture. Among them, the particle size determination process is as follows: transfer an appropriate amount of sample particles to a dry test sample cell, select the dry test mode, and set the air pressure to 0.2 - 0.5 Mpa, thereby realizing the particle size determination of the sample particles. The moisture determination process is as follows: weigh an appropriate amount of sample particles and spread them evenly on a petri dish with a diameter of 60 mm, put them into a rapid moisture analyzer to measure the moisture content, and set the heating temperature to 105 °C, the sample injection volume to 1.0 - 2.0 g, the discrimination time to 50 - 100 s, and measure each sample at least 3 times;

[0069] Step S14: Establish a quantitative calibration model; specifically, the establishment process of the quantitative calibration model includes the following steps:

[0070] Step S141: Collect sample particles during the fluidized bed granulation process. Among them, the particle collection process is as follows: The first robot moves to the vicinity of the fluidized bed under the guidance of the laser detection device. The signal acquisition device aligns with the sampling port position of the fluidized bed and activates the robotic arm. The robotic arm drives the clamping device to move to the sample particle placement position, and the clamping device clamps the sample particles, thereby realizing the collection of sample particles.

[0071] Step S142: Collect hyperspectral images of the sample particles and calculate the average spectrum. Among them, the average spectrum calculation formula is: Among them, is the average reflectance of the current sample at wavelength w, n is the number of pixel points in the hyperspectral image of the current sample, is the average spectrum of the current sample, and then the spectral information of different samples is calculated. The wavelength range for hyperspectral image collection is 390 - 1700 nm, the exposure time is 1 - 5 ms, and the distance between the imaging lens and the sample particles is 20 - 60 cm.

[0072] Step S143: Establish a quantitative calibration model for particle moisture, particle size, and the content of active pharmaceutical ingredients using partial least squares method. Specifically, the partial least squares method can also be combined with appropriate spectral preprocessing methods such as smoothing and differentiation to improve the accuracy of the quantitative calibration model. The evaluation indicators of the quantitative calibration model are: the correlation coefficient Rc of the training set 2 , the correlation coefficient Rp of the test set 2 , the root mean square error RMSEC of the training set, and the root mean square error RMSEP of the test set.

[0073] Step S144: Complete the detection of the fluidized bed granulation process.

[0074] Step S2: Based on the second robot, inspect the pressure and temperature parameters of the fluidized bed equipment;

[0075] Specifically, the inspection process includes the following steps:

[0076] Step S21: Equipment image acquisition. Among them, the image acquisition process is as follows: The robot moves to the vicinity of the fluidized bed under the guidance of the laser detection mechanism, activates the robotic arm, and the robotic arm drives the computer vision mechanism to move to the positions of the target pressure gauge and the target equipment area. The 4K high - definition camera and the infrared thermal imaging camera respectively collect the corresponding images.

[0077] Step S22: Obtain the temperature of each area of the equipment based on the infrared thermal imaging pictures;

[0078] Step S23: Identify the indication of the target pressure gauge. Among them, the identification process of the indication of the target pressure gauge includes the following steps:

[0079] Step S231: Process all the collected pressure gauge images to create a dataset. Specifically, perform data augmentation on the collected images, and use a labeling tool to label the dial, long-end pointer, and mc reference in the processed pressure gauge images. Randomly divide the labeled pressure gauge images into a training set, a test set, and a validation set according to a preset ratio.

[0080] Step S232: Based on the YOLO network model, establish an object detection model for pressure gauge images. Input the processed pressure gauge image dataset and perform model training on the dial, pointer, and reference in the images to obtain a trained pressure gauge image detection model. Specifically, set parameters such as the learning rate and the number of training iterations before training, execute the training code, load the configuration file, and after training is completed, obtain the results of model training, as shown in Table 1 below. In Table 1, mAP represents the area of the closed region formed by using the precision rate and recall rate as the two axes for plotting, and mAP@.5 represents the average mAP with a threshold greater than 0.5.

[0081] Table 1 Training Results of the Pressure Gauge Object Detection Model

[0082] Category P Precision R Recall mAP@.5 All Targets 0.987 1 0.995 Dial 0.975 1 0.995 Benchmark 0.998 1 0.995 Pointer 0.988 1 0.995

[0083] Step S233: Input the image into the trained pressure gauge image detection model, obtain the coordinate position information of the dial, pointer, and reference, and determine the subordinate relationship of the dial, pointer, and reference according to the coordinate values. Specifically, input the pressure gauge image to be detected Figure 10 into the trained pressure gauge image detection model for the positioning of the dial, pointer, and reference, and obtain the four corner coordinates of the dial (164.0, 148.0, 545.0, 533.0), the center coordinate of the dial (354.5, 340.5), the center coordinate of the reference (350.5, 285.0), and the center coordinate of the pointer (290.0, 321.0). Compare the four corner coordinate values of the dial with other coordinate values to determine that they belong to the same pressure gauge;

[0084] Step S234: According to the coordinates of the dial, pointer, and reference obtained in Step S233, calculate the angle between the line connecting the center of the dial and the center of the reference and the line connecting the center of the dial and the center of the pointer. The calculation formula is:

[0085]

[0086] where angle ∈ (-π, π), the center of the dial is (x c , y c ), the center of the pointer is (x b , y b ), the center of the reference is (x a , y a ), and d1, d2 are

[0087] Step S235: Scale conversion. Since the line connecting the reference center and the dial center bisects the scale, the conversion formula is: number = 0.5 + 0.02 × angle / 5.4, where number is the reading of the pressure gauge and angle is the angle between the line connecting the dial center and the reference center and the line connecting the dial center and the pointer center;

[0088] Step S236: Scale printing. Print the scale information onto the camera screen, as Figure 11 shown.

[0089] Step S24: Determine whether the temperature and pressure of the fluidized bed equipment are normal; specifically, identify the pressure gauge in the image captured by the 4K high-definition camera, read the pressure value and compare it with the normal pressure range. If it deviates from the normal range, it is determined as abnormal; read the temperature of the equipment area in the image captured by the infrared thermal imaging camera and compare it with the normal temperature range. If it deviates from the normal range, it is determined as abnormal.

[0090] Step S3: Coordinate the control of the first robot and the second robot to monitor the granulation process.

[0091] Specifically, the coordination control process includes the following steps:

[0092] Step S31: Map location sharing;

[0093] Specifically, the debugging personnel operate one of the robots to move in the workshop in manual mode. The robot senses the workshop environment through the installed lidar and independently draws the workshop map; after the map is drawn and enabled, mark the current positions of other robots on the map to achieve map location sharing.

[0094] Step S32: Detection and patrol abnormal alarm; among them, the abnormal alarm process is: if the detection and patrol results in steps S15 and S24 are abnormal, the abnormal information will be sent to the alarm mechanism of the second robot via the Modbus protocol through the local area network, and the alarm mechanism will alarm through the warning light language and voice. Specifically, when the first robot detects that the particle size or moisture content of the sample particles in the fluidized bed granulation process is abnormal or the second robot patrols that the pressure or temperature of the fluidized bed equipment is abnormal, write the agreed code to the register address 9000. Among them, the code written for too high moisture content is 101, the code written for too low moisture content is 102, the code written for too large particle size is 103, the code written for too small particle size is 104, the code written for too low pressure is 201, the code written for too high pressure is 202, the code written for too high equipment temperature is 203, the code written for too high ambient temperature is 204, etc. When the second robot reads the agreed code in the 9000 register, it will issue an alarm through the warning light language and the voice alarm mechanism.

[0095] Step S33: Regulation of fluidized bed equipment. The regulation process of the fluidized bed equipment is as follows: If the detection and inspection results in steps S15 and S24 are abnormal, the abnormal information is sent to the first robot via the local area network through the Modbus protocol. The first robot executes corresponding tasks according to the type of abnormality, including rotating the specified temperature regulating valve and pressure regulating valve and pressing the equipment emergency stop button to complete autonomous control. Specifically, when the first robot reads the agreed encoding written in step S2 in register 9000, the first robot executes corresponding tasks according to the type of abnormality, including: rotating the temperature regulating valve to increase the temperature when reading 101; rotating the temperature regulating valve to decrease the temperature when reading 102; rotating the pressure regulating valve to increase or decrease the pressure when reading 103; rotating the pressure regulating valve to increase the pressure when reading 104; rotating the pressure regulating valve to increase the pressure when reading 201; pressing the equipment emergency stop button to pause the granulation process when reading 202, 203, 204, etc.

[0096] Example 1

[0097] Based on a method for monitoring the Chinese medicine fluidized bed granulation process based on a robot, taking Ganoderma lucidum spore powder particles as an example, the detection process in step S1 specifically includes the following steps:

[0098] Step S11: Use the first robot to collect sample particles of the Ganoderma lucidum spore powder particle fluidized bed granulation process;

[0099] Specifically, granulation is carried out in one step under the conditions of a pressure of -2.5 to -3.0 Kpa, an inlet air temperature of 60 - 80 °C, and an outlet air temperature not exceeding 60 °C. The detection robot moves to the vicinity of the fluidized bed under the guidance of the lidar, calibrates the sampling port with the servo camera, activates the gripper at the farthest end of the robotic arm, grabs an empty sample bottle, and collects fluidized bed particles. Approximately 10 g of Ganoderma lucidum spore powder particles are collected every 10 - 15 minutes until the granulation process is completed. A total of 57 samples of Ganoderma lucidum spore powder particle fluidized bed particles in 5 batches are collected.

[0100] Step S12: Use the first robot to collect hyperspectral images of the Ganoderma lucidum spore powder particle fluidized bed granulation process particle samples;

[0101] Specifically, the first robot pours the collected fluidized bed particles into the sample inlet of the hyperspectral detection robot, turns on the switch of the hyperspectral detection robot, and uses the hyperspectral imaging system to obtain the hyperspectral images of 57 sample particles. The hyperspectral camera is aligned with the sample directly from above, and there is a halogen light source on each side of the camera. To eliminate the uneven light distribution and dark current generated by the long-term heating of the instrument, as well as the influence of the unstable light source on the image, a standard white calibration plate is collected to obtain a white calibration image, and the image covered by the captured camera lens is stored as a black calibration image. The obtained hyperspectral images are calibrated with black and white calibration images on the image acquisition software. To ensure that the generated images are not distorted and have consistent data sizes, the length of the spectral range is arranged to be 800 pixels. The distance between the imaging lens and the sample is 40 cm, and the exposure time is 1.7 ms. A spectral data cube of 128 wavelengths is obtained at intervals of 6.25 nm in the spectral range of 900 - 1700 nm. The width of the hyperspectral image is 800 pixels, the height is 703 pixels, and there are 128 bands.

[0102] Step S13: Determine the parameters of the sample particles;

[0103] Transfer 5 g of the Ganoderma lucidum spore powder particle sample to a dry test sample cell, select the dry test mode, set the pressure to 0.35 Mpa, and measure the particle size of the particle sample. Weigh 2.0 g of the Ganoderma lucidum spore powder particle sample and put it into a rapid moisture analyzer to measure the moisture content. The heating temperature is 105 °C, the sample injection volume is 2.0 g, the discrimination time is 50 - 100 s, preferably 70 s, and each sample is measured 3 times.

[0104] Step S14: Establish a quantitative calibration model;

[0105] Calculate the average spectrum of the collected hyperspectral images of the samples. The calculation formula is: Where, is the average reflectance of the current sample at wavelength w, n is the number of pixel points of the hyperspectral image of the current sample, is the average spectrum of the current sample, and the spectral information of different samples is calculated therefrom.

[0106] In the training set, a partial least squares algorithm is used to establish a quantitative calibration model for the hyperspectrum of Ganoderma lucidum spore powder particles and moisture content and particle size. After the model is established, the performance of the model is verified with the test set. The model evaluation indicators include the correlation coefficient Rc of the training set 2 , the correlation coefficient Rp of the test set 2 , the root mean square error RMSEC of the training set, and the root mean square error RMSEP of the test set. The closer Rp 2 is to 1, the stronger the prediction effect of the model. The smaller the RMSEC and RMSEP and the closer their values are, the better the prediction performance and robustness of the model.

[0107] Table 2 shows the Rp 2 values and the summary table of RMSEP values for the quantitative calibration model. It can be seen from Table 2 that the Rp 2 values of moisture content, particle size (<10%), and particle size (<50%) in the quantitative calibration model are 0.981, 0.982, and 0.967 respectively.

[0108] Therefore, the quantitative calibration model has the optimal performance, and the prediction results of moisture and particle size in the fluidized bed granulation process of Ganoderma lucidum spore powder particles are reliable, meeting the requirements of rapid detection.

[0109] Table 2 Summary of the results of the particle quantitative calibration model in the fluidized bed granulation process of Ganoderma lucidum spore powder

[0110]

[0111] Step S5: Complete the detection of the fluidized bed granulation process of Ganoderma lucidum spore powder particles. The detection results of moisture and particle size in the fluidized bed process of 5 batches of Ganoderma lucidum spore powder particles described in this embodiment are as Figures 2 - 4 shown, where Figures 2 - 4 the abscissa of both is time, Figure 2 the ordinate of Figure 3 and Figure 4 is the water content, and the ordinate of

[0112] Example 2

[0113] Based on a robot-based monitoring method for the fluidized bed granulation process of traditional Chinese medicine, taking Dendrobium officinale particles as an example, the detection process in step S1 specifically includes the following steps:

[0114] Step S11: Use the first robot to collect sample particles in the fluidized bed granulation process of Dendrobium officinale particles;

[0115] Specifically, granulation is carried out in one step under the conditions of pressure -2.5 to -3.0 Kpa, inlet air temperature 60 - 80 °C, and outlet air temperature not exceeding 60 °C. The detection robot moves to the vicinity of the fluidized bed under the guidance of a lidar, the servo camera calibrates the sampling port, starts the gripper at the farthest end of the robotic arm, grabs an empty sample bottle, and collects fluidized bed particles. Approximately 10 g of Dendrobium officinale particles are collected every 10 - 15 minutes until the granulation process is completed. A total of 86 samples of fluidized bed particles of Dendrobium officinale particles in 5 batches are collected.

[0116] Step S12: Use the first robot to collect hyperspectral images of the sample particles in the fluidized bed granulation process of Dendrobium officinale particles;

[0117] Specifically, the robot pours the collected fluidized bed particles into the sample inlet of the hyperspectral detection robot, turns on the switch of the hyperspectral detection robot, and uses the hyperspectral imaging system to obtain hyperspectral images of 86 sample particles. The hyperspectral camera is aligned with the sample from directly above, and there is a halogen light source on each side of the camera. To eliminate the uneven light distribution and dark current generated by the long-term heating of the instrument, as well as the influence of the unstable light source on the image, a standard white calibration plate is collected to obtain a white calibration image, and the image covered by the captured camera lens is stored as a black calibration image. The obtained hyperspectral images are calibrated with the black and white calibration images on the image acquisition software. To ensure that the generated images are not distorted and have the same data size, the length of the spectral range is arranged to be 800 pixels. The distance between the imaging lens and the sample is 40 cm, and the exposure time is 1.7 ms. A spectral data cube of 128 wavelengths is obtained at intervals of 6.25 nm in the spectral range of 900 - 1700 nm. The width of the hyperspectral image is 800 pixels, the height is 703 pixels, and there are 128 bands.

[0118] Step S13: Determine the sample particle parameters;

[0119] Transfer 5 g of the Dendrobium officinale Kimura et Migo particle sample to a dry test sample cell, select the dry test mode, set the pressure to 0.35 Mpa, and measure the particle size of the particle sample. Weigh 2.0 g of the Dendrobium officinale Kimura et Migo particle sample and put it into a rapid moisture analyzer to measure the moisture content. The heating temperature is 105 °C, the sample injection volume is 2.0 g, and the discrimination time is 70 s. Each sample is measured 3 times.

[0120] Step S14: Establish a quantitative calibration model;

[0121] Calculate the average spectrum of the collected sample hyperspectral images. The calculation formula is: Where, is the average reflectance of the current sample at wavelength w, n is the number of pixel points of the current sample hyperspectral image, is the average spectrum of the current sample, and the spectral information of different samples is calculated from this.

[0122] In the training set, the partial least squares algorithm is used to establish a quantitative calibration model for the hyperspectrum and moisture content, particle size of Ganoderma lucidum spore powder particles. After the model is established, the performance of the model is verified with the test set. The model evaluation indicators include the training set correlation coefficient Rc 2 , the test set correlation coefficient Rp 2 , the root mean square error of the training set RMSEC and the root mean square error of the test set RMSEP. The closer Rp 2 is to 1, the stronger the prediction effect of the model. The smaller the RMSEC and RMSEP and the closer their values are, the better the prediction performance and robustness of the model.

[0123] Table 3 shows the Rp 2 values and the summary table of RMSEP values for the quantitative calibration model. As can be seen from Table 3, the Rp 2 values of moisture content, particle size (<3%), particle size (<6%), particle size (<10%) and particle size (<25%) in the quantitative calibration model are 0.961, 0.903, 0.916, 0.921 and 0.908 respectively.

[0124] Therefore, the quantitative calibration model has the optimal performance, and the prediction results of moisture and particle size in the fluidized bed granulation process of Dendrobium officinale are reliable, meeting the requirements of rapid detection.

[0125] Table 3 Summary of the results of the particle quantitative calibration model in the fluidized bed granulation process of Dendrobium officinale

[0126]

[0127]

[0128] Step S5: Complete the detection of the fluidized bed granulation process of Dendrobium officinale particles. The detection results of moisture and particle size in the fluidized bed process of 5 batches of Dendrobium officinale particles described in this embodiment are as Figures 5 - 9 shown, where Figures 5 - 9 the abscissa of both is time, Figure 5 the ordinate of Figures 6 - 9 is the water content, and the ordinate of

[0129] Example 3

[0130] As Figure 12 shown, an inspection device for the fluidized bed granulation process of traditional Chinese medicine based on a robot includes a robotic arm 1, a computer vision mechanism 2, a traveling mechanism 3 and an alarm mechanism 4; the robotic arm and the alarm mechanism 4 are installed at the upper end of the traveling mechanism 3; the alarm mechanism 4 is composed of a warning light 41 and a voice prompt mechanism 42; the computer vision mechanism 2 is installed at the upper end of the robotic arm 1, and the computer vision mechanism 2 includes a 4K high-definition camera 21 and an infrared thermal imaging camera 22.

[0131] The working principle of this device is as follows: The robot moves to the vicinity of the fluidized bed under the guidance of the traveling mechanism 3, activates the robotic arm 1, and the robotic arm 1 drives the computer vision mechanism 2 to move to the positions of the target pressure gauge and the target equipment area. The 4K high-definition camera 21 and the infrared thermal imaging camera 22 respectively collect corresponding images. If the pressure gauge reading in the image captured by the 4K high-definition camera 21 deviates from the normal range, the temperature of the fluidized bed equipment collected by the infrared thermal imaging camera 22 deviates from the normal range, or information on abnormal moisture content and particle size (deviating from the normal range) of the sample particles in the fluidized bed granulation process is received from other robots, the abnormal information will be sent to the alarm mechanism 4 of the robot via the local area network through the Modbus protocol. The alarm mechanism 4 alarms through the light language of the warning light 41 and the voice prompt mechanism 42. Specifically, the abnormal information is sent by writing a predefined code to the register address 9000. Among them, the code written for too high moisture content is 101, the code written for too low moisture content is 102, the code written for too large particle size is 103, the code written for too small particle size is 104, the code written for too low pressure is 201, the code written for too high pressure is 202, the code written for too high equipment temperature is 203, the code written for too high ambient temperature is 204, etc. When the robot reads the predefined code in the 9000 register, it issues an alarm through the light language of the warning light 41 and the voice prompt mechanism 42.

[0132] Finally, it should be noted that the above content is only used to illustrate the technical solution of the present invention, rather than a limitation on the protection scope of the present invention. Any simple modification or equivalent replacement of the technical solution of the present invention by those of ordinary skill in the art shall not depart from the essence and scope of the technical solution of the present invention.

Claims

1. A robot-based monitoring method for the fluidized bed granulation process of traditional Chinese medicine, characterized in that, It includes the following steps: Step S1: Detect the particle moisture and particle size parameters during the granulation process based on the first robot; Step S2: Inspect the pressure and temperature parameters of the fluidized bed equipment based on the second robot; Among them, the inspection method in Step S2 includes the following steps: Step S21: Collect equipment images; Step S22: Obtain the temperature of each area of the equipment based on the infrared thermal imaging pictures; Step S23: Identify the reading of the target pressure gauge; Step S24: Determine whether the temperature and pressure of the fluidized bed equipment are normal; Specifically, the process of identifying the reading of the target pressure gauge in Step S23 includes the following steps: Step S231: Process all the collected pressure gauge images to make a data set; Step S232: Based on the YOLO network model, establish a target detection model for the pressure gauge images, input the processed pressure gauge image data set, and train the model for the dial, pointer, and reference in the image to obtain a trained pressure gauge image detection model; Step S233: Input the image into the trained pressure gauge image detection model, obtain the coordinate position information of the dial, pointer, and reference, and determine the subordinate relationship of the dial, pointer, and reference according to the coordinate values; Step S234: According to the coordinates of the dial, pointer, and reference obtained in Step S233, calculate the included angle between the line connecting the center of the dial to the center of the reference and the line connecting the center of the dial to the center of the pointer. The calculation formula is: where angle ∈ (-π, π), the center of the dial is (x c , y c ), the center of the pointer is (x b , y b ), the reference center is (x a , y a ), and d1, d2 are Step S235: Scale conversion. Since the line connecting the center of the reference to the center of the dial bisects the scale, the conversion formula is: number = 0.5 + 0.02 × angle / 5.4, where number is the reading of the pressure gauge and angle is the included angle between the line connecting the center of the dial to the center of the reference and the line connecting the center of the dial to the center of the pointer; Step S236: Print the scale information onto the camera screen; Step S3: Cooperatively control the first robot and the second robot to monitor the granulation process. Among them, the cooperative control process in Step S3 includes the following steps: Step S31: Share the map position; Step S32: Detect and patrol for abnormal alarms; Specifically, the abnormal alarm process in Step S32 is as follows: If the detection and patrol results in Steps S15 and S24 are abnormal, the abnormal information will be sent to the alarm mechanism of the second robot via the Modbus protocol through the local area network, and the alarm mechanism will give an alarm through the warning light signal and voice; Step S33: Regulate the fluidized bed equipment; Specifically, the process of regulating the fluidized bed equipment is as follows: If the detection and patrol results in Steps S15 and S24 are abnormal, the abnormal information will be sent to the first robot via the Modbus protocol through the local area network, and the first robot will execute corresponding tasks according to the abnormal type, including rotating the specified temperature regulating valve and pressure regulating valve and pressing the emergency stop button of the equipment to complete autonomous control.

2. The method for monitoring the traditional Chinese medicine fluidized bed granulation process based on a robot according to claim 1, wherein, The detection process in Step S1 includes the following steps: Step S11: Collect sample particles; Step S12: Collect the hyperspectral images of the sample particles; Step S13: Measure the mass parameters of the sample particles; Step S14: Establish a quantitative calibration model; Among them, the process of establishing the quantitative calibration model includes the following steps: Step S141: Collect sample particles during the fluidized bed granulation process; Step S142: Collect hyperspectral images of the sample particles and calculate the average spectrum; Step S143: Measure the particle size and moisture content of the sample particles; Step S144: Establish a quantitative calibration model for particle moisture content, particle size, and the content of the pharmaceutically active substance using partial least squares method; Step S145: Complete the detection of the fluidized bed granulation process.

3. The monitoring method for the traditional Chinese medicine fluidized bed granulation process based on a robot according to claim 2, characterized in that In the particle collection process of Step S11: The first robot moves to the fluidized bed under the guidance of the laser detection device. The signal acquisition device is aligned with the sampling port position of the fluidized bed, and the robotic arm is activated. The robotic arm drives the clamping device to move to the position where the sample particles are placed, and the clamping device clamps the sample particles, thereby achieving the collection of the sample particles.

4. The monitoring method for the traditional Chinese medicine fluidized bed granulation process based on a robot according to claim 2, wherein In the hyperspectral image collection process of Step S12: The sample particles clamped by the clamping device are placed into the hyperspectral detection device through the injection port. The switch is activated, and the hyperspectral data of the sample particles are obtained through the imaging system and transmitted to the remote server.

5. A method for monitoring the Chinese medicine fluidized bed granulation process based on a robot according to claim 2, characterized in that, The average spectrum calculation formula in step S142 is as follows: Among them, is the average reflectance of the current sample at band w, n is the number of pixel points in the hyperspectral image of the current sample, is the average spectrum of the current sample, and then the spectral information of different samples is calculated.

6. A monitoring device adopting the monitoring method for the traditional Chinese medicine fluidized bed granulation process according to any one of claims 1-5, characterized in that, It includes a robotic arm, a computer vision mechanism, a walking mechanism, and an alarm mechanism; The upper end of the walking mechanism is equipped with a robotic arm and an alarm mechanism; The alarm mechanism consists of a warning light and a voice prompt mechanism; The upper end of the robotic arm is equipped with a computer vision mechanism.

Images