Furniture production control method, terminal device and storage medium

Abstract

The application is suitable for the technical field of furniture production, and provides a furniture production control method, a terminal device and a storage medium. The method comprises the following steps: determining a plurality of processing devices on a production line of furniture; collecting original image data of a part and collecting original point cloud data of the part; screening target image data representing the part from the original image data and screening target point cloud data representing the part from the original point cloud data; reading a configuration file of the processing device; loading a quality inspection program corresponding to a quality inspection label according to the similarity of the processing device on the part; starting the quality inspection program, performing a quality inspection operation on the part according to the target image data and / or the target point cloud data, and obtaining quality information; if a quality inspection instruction is received when the quality information is displayed on the part model, processing the part according to the quality inspection instruction and processing the quality inspection program according to the quality information and the quality inspection instruction. The above method can effectively improve the yield of furniture production.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of furniture production, and particularly relates to a furniture production control method, a terminal device and a storage medium. BACKGROUND

[0002] With the development of economy, furniture gradually becomes a fast-moving consumer product, and furniture presents a change trend of small scale and rapid model change.

[0003] Therefore, the production equipment of furniture is continuously fine-tuned to adapt to the change trend of furniture. Since fine-tuning has a certain impact on the stability of the production equipment, it is easy to cause certain defects in the furniture during the production of the furniture.

[0004] At present, during furniture production, technical personnel are mainly responsible for quality inspection of the furniture and checking whether the furniture has defects.

[0005] However, some defects are relatively small, so that some defects are missed, and when the model of the furniture is changed, the defects also change, and the corresponding quality inspection method also changes, so that the quality inspection has a certain learning cost, which aggravates the defect missing, reduces the efficiency of furniture production, and results in a low yield of furniture production. SUMMARY

[0006] Therefore, the application provides a furniture production control method, a terminal device and a storage medium to improve the yield of furniture production.

[0007] A first aspect of an embodiment of the application provides a furniture production control method, comprising:

[0008] determining a plurality of processing equipment on a production line of furniture; each of the processing equipment is used for processing a part of the furniture, and each of the processing equipment is configured with a trigger, a camera and a radar;

[0009] when the processing equipment delivers the part that has completed processing to a track and the part activates the trigger, simultaneously calling the camera to collect original image data of the part and calling the radar to collect original point cloud data of the part;

[0010] comparing the original image data and the original point cloud data with each other to screen target image data representing the part from the original image data and screen target point cloud data representing the part from the original point cloud data;

[0011] reading a configuration file of the processing equipment; the configuration file records one or more quality inspection labels;

[0012] The quality inspection program corresponding to the quality inspection label is loaded according to similarity of the machining equipment in machining the part;

[0013] The quality inspection program is started, and a quality inspection operation is performed on the part according to the target image data and / or the target point cloud data to obtain quality information;

[0014] A three-dimensional part model is constructed according to the target image data and the target point cloud data of the part;

[0015] If a quality inspection instruction is received when the quality information is displayed on the part model, the part is processed according to the quality inspection instruction, and the quality inspection program is processed according to the quality information and the quality inspection instruction.

[0016] A second aspect of the embodiment of the application provides a furniture production control device, comprising:

[0017] A machining equipment determination module is configured to determine a plurality of machining equipment on a production line of furniture, each of the machining equipment being configured to machine a part of the furniture, and each of the machining equipment being provided with a trigger, a camera and a radar;

[0018] An original data acquisition module is configured to simultaneously call the camera to acquire original image data of the part and call the radar to acquire original point cloud data of the part when the machining equipment delivers the part machined to a track and the part activates the trigger;

[0019] A target data screening module is configured to compare the original image data and the original point cloud data with each other to screen target image data representing the part from the original image data and screen target point cloud data representing the part from the original point cloud data;

[0020] A configuration file reading module is configured to read a configuration file of the machining equipment, and the configuration file records one or more quality inspection labels;

[0021] A quality inspection program loading module is configured to load a quality inspection program corresponding to the quality inspection label according to similarity of the machining equipment in machining the part;

[0022] A part quality inspection module is configured to start the quality inspection program, perform a quality inspection operation on the part according to the target image data and / or the target point cloud data, and obtain quality information;

[0023] A part model construction module is configured to construct a three-dimensional part model according to the target image data and the target point cloud data of the part;

[0024] The quality inspection processing module is configured to, if a quality inspection instruction is received when the quality information is displayed on the part model, process the part according to the quality inspection instruction, and process the quality inspection procedure according to the quality information and the quality inspection instruction.

[0025] The third aspect of the embodiment of the present application provides a terminal device, which comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the furniture production control method according to the first aspect.

[0026] The fourth aspect of the embodiment of the present application provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the furniture production control method according to the first aspect.

[0027] The fifth aspect of the embodiment of the present application provides a computer program product, which, when executed on a computer, enables the computer to execute the furniture production control method according to the first aspect.

[0028] Compared with the prior art, the embodiment of the present application has the following beneficial effects:

[0029] In the embodiment, a plurality of processing devices are determined on a production line of furniture; each processing device is used for processing a part of the furniture, and each processing device is configured with a trigger, a camera, and a radar; when the processing device delivers the part processed to a track and the part activates the trigger, the camera is simultaneously called to collect original image data of the part, and the radar is simultaneously called to collect original point cloud data of the part; the original image data and the original point cloud data are compared with each other to filter target image data representing the part from the original image data and filter target point cloud data representing the part from the original point cloud data; a configuration file of the processing device is read; one or more quality inspection tags are recorded in the configuration file; a quality inspection program corresponding to the quality inspection tag is loaded according to the similarity of the processing device in processing the part; the quality inspection program is started, and a quality inspection operation is performed on the part according to the target image data and / or the target point cloud data to obtain quality information; a three-dimensional part model of the part is constructed according to the target image data and the target point cloud data; if a quality inspection instruction is received when the quality information is displayed on the part model, the part is processed according to the quality inspection instruction, and the quality inspection program is processed according to the quality information and the quality inspection instruction. The embodiment combines image data in vision and point cloud data in touch to construct a quality inspection framework of the furniture, realizes a universal furniture quality inspection scheme, decouples the processing business and the quality inspection logic of the furniture, flexibly adjusts the programs and the tags therebetween under the quality inspection framework of the furniture, can quickly follow up the quality inspection scheme when the production device is slightly adjusted, reduces the learning cost of quality inspection, and the image data in vision and the point cloud data in touch can touch the small defects of the furniture, the accuracy of quality inspection is high, the situation of missed inspection is greatly reduced, and thus the efficiency of furniture production is effectively improved and the yield of furniture production is improved. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating laborious work.

[0031] Figure 1 is a schematic diagram of a furniture production control method provided by the embodiment of the present application;

[0032] Figure 2 is a schematic diagram of a quality inspection network provided by the embodiment of the present application;

[0033] Figure 3 is a schematic diagram of a furniture production control device provided by the embodiment of the present application;

[0034] Figure 4 is a schematic diagram of a terminal device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0035] In the following description, for purposes of explanation and not limitation, specific details are set forth such as particular architectures, technologies, techniques, etc. in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0036] The technical solutions of the present application are described below through specific embodiments.

[0037] Referring to Figure 1 , a schematic diagram of a furniture production control method provided by an embodiment of the present application is shown, which can specifically include the following steps:

[0038] Step 101, determining a plurality of processing devices on a production line of furniture.

[0039] In the present embodiment, one or more production lines are arranged in a production workshop, wherein one production line can generate one or more types of furniture, and a plurality of fully automatic or semi-automatic processing devices are provided on the production line, each of which is used for processing parts of furniture, such as pressing, cutting, punching, polishing, painting, etc.

[0040] Among them, the parts of furniture refer to the components of furniture, for example, the parts of a bed include a headboard, a bed board, a bed beam, a bed leg, etc.

[0041] Each processing device is configured with a trigger, a camera and a radar, the trigger is electrically connected with a controller, and the controller is electrically connected with the camera and the radar respectively.

[0042] Among them, the controller includes a PLC (Programmable Logic Controller) and the like, the trigger can be activated by infrared, pressure and the like, the camera includes an RGB (Red Green Blue) camera and the like, and the radar includes a low-line laser radar and the like.

[0043] The camera can provide visual image data, and the radar can provide tactile point cloud data, both of which can perceive the parts of furniture from different dimensions to provide a data basis for general quality inspection.

[0044] Step 102, when the processing device delivers the part to be completed to a track and the part activates the trigger, simultaneously calling the camera to collect original image data of the part and calling the radar to collect original point cloud data of the part.

[0045] In a system such as EAP (Equipment Automation Programming), the status of each processing equipment can be monitored. If a processing equipment completes processing of a certain part of furniture, the processing equipment delivers the part to a track for quality inspection. At this time, when the part passes through a trigger of the processing equipment, the trigger is activated, indicating that the part has entered the sensing range of a camera and a radar. The trigger sends a trigger signal to a controller, and the controller simultaneously sends an enable signal to the camera and the radar. The camera acquires image data in its sensing range (including the part) according to the enable signal, which is recorded as original image data. The radar acquires point cloud data in its sensing range (including the part) according to the enable signal, which is recorded as original point cloud data.

[0046] Step 103, compare the original image data and the original point cloud data with each other to filter target image data representing the part from the original image data and target point cloud data representing the part from the original point cloud data.

[0047] In actual applications, the original image data and the original point cloud data are data sensed from different dimensions for the same part. The contents of the two are the same, but there are certain differences between the data. Therefore, the original image data and the original point cloud data can be compared with each other to find the same places and the different places between the two. On the one hand, the pixels representing the part are filtered from the original image data as target image data. On the other hand, the (radar) points representing the part are filtered from the original point cloud data as target point cloud data.

[0048] Then, the target image data and the target point cloud data pay more attention to the part itself, excluding the influence of the background, and provide the object of attention for general quality inspection.

[0049] In an embodiment of the present application, step 103 can include the following steps:

[0050] Step 1031, read a conversion matrix for calibration between the camera and the radar.

[0051] In the present embodiment, when the camera and the radar are installed, joint calibration can be performed on the camera and the radar to obtain a conversion matrix between the two. The conversion matrix represents the conversion relationship (such as translation relationship, rotation relationship, etc.) between the coordinate system of the camera and the coordinate system of the radar.

[0052] Step 1032, project the original point cloud data into the original image data according to the conversion matrix to obtain reorganized image data.

[0053] In the embodiment, each (radar) point of the original point cloud data is multiplied by the transfer matrix to obtain each (radar) point in the coordinate system of the camera, so as to project each (radar) point of the original point cloud data into the original image data. At this time, the original image data contains both the information (such as RGB) of the pixel points and the information (such as reflection intensity, spatial coordinates, etc.) of the (radar) points. For the convenience of distinction, the original image data is recorded as the reorganized image data.

[0054] In step 1033, the reference image data collected by the camera is called when the processing equipment is not processing the parts.

[0055] Generally, the position of the processing equipment and the track thereof is fixed. Therefore, when the processing equipment is not processing the parts, the track does not have the parts thereon. At this time, the reference image data collected by the camera is called. The reference image data at this time is the pure background.

[0056] In step 1034, the difference image data between the original image data and the reference image data is calculated as the first candidate image data.

[0057] In the embodiment, the difference image data between the original image data and the reference image data can be calculated, which is recorded as the first candidate image data. At this time, the background can be effectively eliminated, and the main content in the first candidate image data is the parts.

[0058] Since the track of the processing equipment is continuously running, and the running thereof is usually periodic, some background data may exist in the difference image data. Therefore, when the processing equipment is not processing the parts, multiple frames of reference image data collected by the camera can be called to cover the running period of the track of the processing equipment.

[0059] If there are multiple frames of reference image data, the difference image data between the original image data and each frame of reference image data can be calculated.

[0060] The areas of the foreground data in each frame of difference image data are counted, and the areas of the foreground data in each frame of difference image data are compared.

[0061] The difference image data with the smallest area of the foreground data is selected to obtain the first candidate image data. The influence of the periodic running of the track of the processing equipment can be effectively reduced, and the operation is simple and fast.

[0062] In step 1035, the reference surface representing the track is constructed according to the original point cloud data.

[0063] In actual application, the parts are located above the track, and thus, a clustering algorithm such as DBSCAN (Density Based Spatial Clustering of Applications with Noise) can be used to cluster the original point cloud data to construct the datum plane representing the track.

[0064] Step 1036, filtering data above the datum plane in the reorganized image data to obtain second candidate image data.

[0065] In this embodiment, each data (pixel point, (radar) point) in the reorganized image data is compared with the datum plane, and if the data is above the datum plane, the data can be filtered to form the second candidate image data.

[0066] At this time, the periodic operation of the track of the processing equipment has a certain impact on the datum plane, so that the second candidate image data can have some background data.

[0067] Step 1037, taking the intersection of the first candidate image data and the second candidate image data to obtain feature image data.

[0068] The impact of the periodic operation of the track of the processing equipment on the first candidate image data and the second candidate image data is different, that is, there can be a large difference in the background data contained in the first candidate image data and the second candidate image data.

[0069] Since the first candidate image data has pixel points and the second candidate image data has pixel points, the intersection of the first candidate image data and the second candidate image data can be directly taken to obtain feature image data (containing pixel points and (radar) points), which can effectively exclude background data and obtain relatively pure foreground data.

[0070] Step 1038, determining that the pixel points in the feature image data form target image data representing the parts, and the points in the feature image data form target point cloud data representing the parts.

[0071] In the case where the feature image data is relatively pure foreground data, on the one hand, the pixel points in the feature image data can form target image data representing the parts, and on the other hand, the (radar) points in the feature image data can form target point cloud data representing the parts.

[0072] Step 104, reading a configuration file of the processing equipment.

[0073] In the embodiment, the configuration file of the machining device can be set according to the requirement of quality inspection of the parts machined by the machining device, wherein the configuration file records one or more quality inspection tags, and the quality inspection tag is information identifying the same type of quality inspection procedure, and the quality inspection procedure is a procedure of quality inspection of the parts machined by the machining device.

[0074] When the machining device is micro-adjusted or replaced with a new part, the technician can select a suitable quality inspection procedure according to the actual situation, write the quality inspection tag of the quality inspection procedure into the configuration file, realize hot update of the configuration file, decouple the business (i.e. machining of parts by the machining device) from the underlying logic (i.e. quality inspection procedure), and let the technician focus more on developing general quality inspection procedures.

[0075] Step 105: loading the quality inspection procedure corresponding to the quality inspection tag according to the similarity of the machining device in machining the parts.

[0076] In actual application, a production workshop is a weak network environment, and there is a relatively strict access mechanism between the internal local area network and the external Internet to ensure the network security of the production workshop. The stability of downloading data from the external Internet is poor, and the downloading interruption is prone to occur.

[0077] Considering that the machining device has certain similarity in machining the parts, the quality inspection procedure has certain generality, therefore, the quality inspection procedure corresponding to the quality inspection tag used by other machining devices can be loaded according to the similarity of the machining device in machining the parts, so as to improve the efficiency of loading the quality inspection procedure corresponding to the quality inspection tag.

[0078] In an embodiment of the present application, step 105 can include the following steps:

[0079] Step 1051: reading the process information of the parts machined by the machining device set for the machining device.

[0080] The process information of the parts machined by the machining device set for the machining device is read from the production plan of the furniture, the instruction manual of the machining device and the like.

[0081] Step 1052: screening a plurality of keywords from the process information.

[0082] In the embodiment, a natural language processing algorithm such as TextRank can be used to screen a plurality of keywords from the process information.

[0083] Step 1053: converting the plurality of keywords into a process vector of the machining device.

[0084] In the embodiment, one-hot, word2vec and the like algorithm can be used to vectorize the plurality of keywords, so as to convert them into a process vector of the machining device.

[0085] Step 1054, calculating similarity between the process vector of the current processing device and the process vectors of other processing devices.

[0086] In this embodiment, the similarity between the process vector of the current processing device and the process vectors of other processing devices (i.e., all processing devices except the current processing device) can be calculated using cosine angle algorithm or the like.

[0087] Step 1055, if the similarity is greater than or equal to a preset first threshold, determining the other processing device as a candidate device.

[0088] The similarity is compared with the preset first threshold. If the similarity is greater than or equal to the first threshold, it indicates that the similarity between the process vector of the current processing device and the process vectors of other processing devices is high, and the other processing device is determined as a candidate device.

[0089] Step 1056, if the candidate device has configured the quality inspection program corresponding to the quality inspection label, and the running time of the quality inspection program corresponding to the quality inspection label on the candidate device exceeds a preset second threshold, loading the quality inspection program corresponding to the quality inspection label from the candidate device.

[0090] If the candidate device has configured the quality inspection program corresponding to the quality inspection label, the running time of the quality inspection program corresponding to the quality inspection label on the candidate device can be counted, and the running time of the quality inspection program corresponding to the quality inspection label on the candidate device is compared with the preset second threshold.

[0091] If the running time of the quality inspection program corresponding to the quality inspection label on the candidate device exceeds the second threshold, it indicates that the running time of the quality inspection program corresponding to the quality inspection label on the candidate device is long, and the quality inspection program corresponding to the quality inspection label belongs to a stable quality inspection program. Therefore, the quality inspection program corresponding to the quality inspection label is loaded from the candidate device, and the quality inspection program corresponding to the quality inspection label is configured to the current processing device.

[0092] In a specific implementation, if the current processing device does not configure the quality inspection program corresponding to the quality inspection label, the quality inspection program corresponding to the quality inspection label with the highest similarity can be directly loaded from the candidate device.

[0093] If the current processing device has configured the quality inspection program corresponding to the quality inspection label, the first version information of the quality inspection program corresponding to the quality inspection label in the current processing device and the second version information of the quality inspection program corresponding to the quality inspection label in the candidate device are respectively queried.

[0094] If the first version information is lower than the second version information, and the upgrading of the quality inspection procedure corresponding to the quality inspection label in the current processing device to the second version information is not prohibited, one or more differential upgrade packages between the first version information and the second version information are downloaded from the candidate device, and one differential upgrade package is used to upgrade one version.

[0095] The quality inspection procedure corresponding to the quality inspection label in the current processing device is upgraded using the differential upgrade package, i.e., the quality inspection procedure corresponding to the quality inspection label in the current processing device is upgraded from the first version information to the second version information.

[0096] If the first version information is higher than the second version information, and / or the upgrading of the quality inspection procedure corresponding to the quality inspection label in the current processing device to the second version information is prohibited (i.e., the quality inspection procedure corresponding to the quality inspection label in the current processing device does not meet the preset performance condition when upgraded to the second version information, and is not suitable for the quality inspection of the current part), the quality inspection procedure corresponding to the quality inspection label configured for the candidate device is ignored.

[0097] In step 106, the quality inspection procedure is started, and the quality inspection operation is performed on the part according to the target image data and / or the target point cloud data to obtain the quality information.

[0098] In actual applications, the quality inspection procedure can be loaded using a thread, and the quality inspection operation is performed on the part according to the target image data and / or the target point cloud data to obtain the quality information of the part.

[0099] On one hand, the quality inspection operation supported by the single target image data includes the detection of the coverage and uniformity of paint spraying, foreign matter detection, and the like.

[0100] On the other hand, the quality inspection operation supported by the single target point cloud data includes the size detection of a contour and the arc detection of a contour, and the like.

[0101] On the other hand, the quality inspection operation supported by the single target point cloud data includes the size detection of a contour and the arc detection of a contour, and the like.

[0102] In an embodiment of the present application, step 106 can include the following steps:

[0103] In step 10611, the quality inspection procedure is started.

[0104] In the embodiment, the quality inspection procedure includes a quality inspection network constructed and trained based on deep learning and some other quality inspection logic, and when the quality inspection procedure is started, the quality inspection network is started.

[0105] For example, the quality inspection procedure includes a quality inspection network constructed and trained based on deep learning, and when the quality inspection procedure is started, the quality inspection network is started. Figure 2As shown, the quality inspection network includes a first feature structure Feature_1, a second feature structure Feature_2, a first head structure Head_1, and a second head structure Head_2.

[0106] In step 10612, the feature image data composed of the target image data and the target point cloud data is converted into gray image data.

[0107] In this embodiment, the feature image data composed of the target image data and the target point cloud data is read, and the RGB color value or the like of the pixel point in the feature image data is converted into a gray value under the condition of preserving the (radar) points, so as to obtain the gray image data.

[0108] In step 10613, the gray image data is input into the first feature structure to extract the public feature.

[0109] In this embodiment, the gray image data is input into the first feature structure Feature_1 to extract the public feature applicable to both the first head structure Head_1 and the second head structure Head_2.

[0110] In one design, as shown in Figure 2 The first feature structure Feature_1 includes a first convolution block ConvBlock_1 and a second convolution block ConvBlock_2, and both the first convolution block ConvBlock_1 and the second convolution block ConvBlock_2 encapsulate a layer structure related to convolution, so as to realize down-sampling of features, for example, a convolution layer, a BN (Batch Normalization) layer, a ReLU (Rectified Linear Unit) layer, a pooling layer, and the like.

[0111] Correspondingly, the public feature includes a first basic feature and a second basic feature.

[0112] Then, the gray image data is input into the first convolution block ConvBlock_1 to perform a convolution operation according to the structure thereof, so as to obtain the first basic feature.

[0113] The first basic feature is input into the second convolution block ConvBlock_2 to perform a convolution operation according to the structure thereof, so as to obtain the second basic feature.

[0114] In step 10614, the public feature is input into the first head structure to segment the crack data on the part.

[0115] In this embodiment, the public feature is input into the first head structure Head_1 to complete a semantic segmentation operation, so as to segment out data with a semantic of a crack on a part, which is recorded as crack data, that is, the crack data represents a crack appearing on the part.

[0116] In the process of furniture making, cracks may occur in various processes such as pressing, cutting, punching, polishing, etc. due to material (especially wood), process (excessive temperature, excessive pressure, etc.), and other factors. Cracks are a common quality problem.

[0117] In one design, as shown in Figure 2 The first head structure Head_1 includes a first deconvolution block DeConvBlock_1 and a second deconvolution block DeConvBlock_2, both of which encapsulate layers related to deconvolution to achieve feature upsampling, such as deconvolution layers, BN layers, ReLU, pooling layers, etc.

[0118] Further, the structure between the first deconvolution block DeConvBlock_1 and the second deconvolution block DeConvBlock_2 can be the same or different, and the present embodiment does not limit this.

[0119] Then, the second basic feature is input into the first deconvolution block DeConvBlock_1 to perform deconvolution operation according to its structure, and the first reconstructed feature is obtained.

[0120] The first reconstructed feature and the first basic feature are spliced Concat to form the second reconstructed feature, realizing skip connection to improve the information amount of the feature.

[0121] The second reconstructed feature is input into the second deconvolution block DeConvBlock_2 to perform deconvolution operation according to its structure, and the crack data on the part is obtained.

[0122] Step 10615, input the public domain feature into the second feature structure to convert it into a private domain feature.

[0123] In the present embodiment, the public domain feature is input into the second feature structure Feature_2 for dimension reduction or adjustment, and is converted into a private domain feature to better adapt to the task of target detection.

[0124] In one design, as shown in Figure 2As shown, the second feature Feature_2 includes a residual network ResNet, a pooling layer Pooling, and a multilayer perceptron (MLP). Generally, the residual network ResNet is a shallow residual structure, such as 18 layers, 34 layers, etc., and the pooling layer Pooling is an ASPP (Atrous Spatial Pyramid Pooling). When facing public features with a large amount of information, a larger receptive field is obtained to extract more context information.

[0125] Then, the first base feature and the second base feature are concatenated as a third base feature.

[0126] The third base feature is input into the residual network ResNet to extract a first intermediate feature according to the structure of the residual network ResNet.

[0127] The first intermediate feature is input into the pooling layer Pooling to perform a pooling operation according to the structure of the pooling layer Pooling to obtain a second intermediate feature.

[0128] The second intermediate feature is input into the multilayer perceptron MLP to be mapped as a private domain feature.

[0129] In step 106, the private domain feature is input into the second head structure to detect the defect data on the part.

[0130] In this embodiment, the private domain feature is input into the second head structure Head_2 to complete the target detection operation, so as to detect the data with the semantic of the defect on the part, which is recorded as defect data. That is, the defect data represents the defect on the part.

[0131] In the process of furniture manufacturing, due to factors such as material (especially wood), process (such as blade shaking, etc.), etc., defects such as concave and warped edges may occur in processes such as pressing, cutting, punching, polishing, etc. Defects are a common quality problem.

[0132] In this embodiment, the two common quality problems of cracks and defects are packaged as a task of a quality inspection network. Compared with using independent deep learning models to detect cracks and defects respectively, the quality inspection network has higher universality, lower resource occupation, and lower probability of runtime error, and is suitable for production workshops with limited local area network resources. In addition, managing a single quality inspection network in a production workshop can improve the simplicity of management.

[0133] In one design, as shown in FIG. 7, the quality inspection network includes a first head structure Head_1 and a second head structure Head_2. Figure 3As shown, the second head structure Head_2 includes a third convolution block ConvBlock_3 and a plurality of fully connected layers (FC), and the third convolution block ConvBlock_3 encapsulates a plurality of layers related to convolution, such as a convolution layer, a BN (Batch Normalization) layer, a ReLU (Rectified Linear Unit) layer, a pooling layer, etc., to realize down-sampling of features.

[0134] Further, the structures among the first convolution block ConvBlock_1, the second convolution block ConvBlock_2 and the third convolution block ConvBlock_3 can be the same or different, which is not limited in the embodiment.

[0135] Then, the private domain features are input into the third convolution block ConvBlock_3 to perform convolution operation according to the structure thereof, to obtain target features.

[0136] The target features are sequentially input into the plurality of fully connected layers FC to map the missing data on the parts according to the structure thereof.

[0137] In the embodiment, the quality inspection network realizes modular design, each module is responsible for completing a specific task, each module can be designed, trained and optimized independently, and can reuse third-party pre-trained structures, or can be fine-tuned and reconstructed according to the characteristics of furniture, thereby improving the flexibility and interpretability of the quality inspection network, and facilitating the debugging and optimization of the quality inspection network according to the fine adjustment of furniture production.

[0138] Step 10617, a first abnormality index is counted for the crack data, and a second abnormality index is counted for the missing data.

[0139] In the embodiment, the first abnormality index is counted for the crack data from multiple dimensions, such as the length and width of the minimum circumscribed rectangle of the crack data, the actual length of the crack data, the actual maximum width of the crack data, etc., and the second abnormality index is counted for the missing data from multiple dimensions, such as the length and width of the missing data, the actual area of the missing data, the shortest distance between the missing data and the edge, etc.

[0140] Step 10618, the first quality inspection requirement information and the second quality inspection requirement information set for the parts are queried.

[0141] In the embodiment, the first quality inspection requirement information can be set for the cracks on each part in multiple dimensions according to the production design requirements of the furniture, and the second quality inspection requirement information can be set for the defects on each part in multiple dimensions.

[0142] Step 10619, if the first abnormal index meets the first quality inspection requirement information, and the second abnormal index meets the second quality inspection requirement information, the quality information of the part is determined to be quality qualified.

[0143] In this embodiment, the first abnormal index is compared with the corresponding first quality inspection requirement information, and the second abnormal index is compared with the corresponding second quality inspection requirement information.

[0144] If the first abnormal index meets the corresponding first quality inspection requirement information, and the second abnormal index meets the corresponding second quality inspection requirement information, it means that the cracks on the part meet the quality inspection requirements, and the defects on the part meet the quality inspection requirements. Therefore, the quality information of the part is determined to be quality qualified, and the part is allowed to enter the next process.

[0145] Step 10620, if the first abnormal index does not meet the first quality inspection requirement information, and / or the second abnormal index does not meet the second quality inspection requirement information, the crack data and the defect data are set as the quality information of the part.

[0146] If the first abnormal index does not meet the corresponding first quality inspection requirement information, and / or the second abnormal index does not meet the corresponding second quality inspection requirement information, it means that the cracks on the part do not meet the quality inspection requirements, and the defects on the part do not meet the quality inspection requirements. Therefore, the quality information of the part is determined to be quality qualified, and the part is suspended from entering the next process.

[0147] Step 107, constructing a three-dimensional part model of the part according to the target image data and the target point cloud data.

[0148] In this embodiment, a high-precision three-dimensional part model can be constructed by combining the target image data and the target point cloud data.

[0149] In specific implementation, on one hand, a mesh, voxelization, or other algorithm can be called to use the target point cloud data to construct a three-dimensional model skeleton structure of the part, and the model skeleton structure can be optimized by smoothing, filling holes, etc.

[0150] On the other hand, texture maps can be extracted from the target image data, such as diffuse maps, specular maps, normal maps, roughness maps, etc.

[0151] The texture map model skeleton structure is input into a renderer under a Pytorch3D or Redner framework, the texture map is rendered into the model skeleton structure, and a three-dimensional part model is obtained, improving the details and realism of the model.

[0152] If the quality inspection instruction is received when the quality information is displayed on the part model, the part is processed according to the quality inspection instruction, and the quality inspection procedure is processed according to the quality information and the quality inspection instruction.

[0153] In the embodiment, the part model can be displayed, and the quality information can be displayed on the part model using visualized elements. The technician can adjust the view by zooming, panning, rotating, and the like, so as to check the part and the quality information thereof.

[0154] For example, when the quality information is that the quality is qualified, a green quality inspection mark is displayed on the part model; when the quality information is crack data, defect data, and the quality is unqualified, the crack data and the defect data are highlighted in red, and a red quality inspection mark is displayed, and the like.

[0155] When the technician completes the checking, the quality inspection instruction can be triggered. At this time, the part can be processed according to the indication of the quality inspection instruction.

[0156] If the quality inspection instruction indicates that the quality inspection is successful, the part is allowed to be transported to the next processing equipment on the production line for processing. The order between the processing equipment can be set when the plan for arranging the furniture production is arranged.

[0157] If the quality inspection instruction indicates that the quality inspection fails, the part is removed from the production line, and the part is prohibited from being transported to the next processing equipment for processing. At this time, the production engineer can repair the part. If the repair fails, the part is scrapped.

[0158] In addition, the quality information and the quality inspection instruction can have the same quality inspection conclusion, or can have opposite quality inspection conclusions. Then, the performance of the quality inspection procedure can be evaluated according to the quality information and the quality inspection instruction, so as to process the quality inspection procedure.

[0159] In a specific implementation, if the quality inspection procedure is upgraded from first version information to second version information, the quality information and the quality inspection instruction are compared to calculate a plurality of statistical indicators, such as recall rate, accuracy rate, precision, F1 value, and the like.

[0160] The plurality of statistical indicators are compared with a preset performance condition.

[0161] If the plurality of statistical indicators satisfy the preset performance condition, it indicates that the performance of the quality inspection procedure under the second version information meets the requirement. It is determined that the quality inspection procedure under the second version information is effective for the quality inspection operation of the processing equipment, and the quality inspection procedure under the second version information is maintained to perform the quality inspection operation of the processing equipment.

[0162] If the plurality of statistical indicators do not meet the preset performance condition, indicating that the performance of the quality inspection procedure under the second version information does not meet the requirement, it is determined that the quality inspection procedure under the second version information is invalid for the quality inspection operation of the processing equipment, the quality inspection procedure is rolled back from the second version information to the first version information, and the quality inspection procedure under the first version information with more stable performance is used to perform the quality inspection operation of the processing equipment.

[0163] In the embodiment, a plurality of processing equipment is determined on a production line of furniture; each processing equipment is used for processing a part of furniture, and each processing equipment is configured with a trigger, a camera and a radar; when the processing equipment delivers the part processed to a track and the part activates the trigger, the camera is called to collect original image data of the part and the radar is called to collect original point cloud data of the part at the same time; the original image data and the original point cloud data are compared with each other to filter target image data representing the part from the original image data and filter target point cloud data representing the part from the original point cloud data; a configuration file of the processing equipment is read; one or more quality inspection labels are recorded in the configuration file; a quality inspection procedure corresponding to the quality inspection label is loaded according to the similarity of the processing equipment in processing the part; the quality inspection procedure is started, a quality inspection operation is performed on the part according to the target image data and / or the target point cloud data to obtain quality information; a three-dimensional part model of the part is constructed according to the target image data and the target point cloud data; if a quality inspection instruction is received when the quality information is displayed on the part model, the part is processed according to the quality inspection instruction, and the quality inspection procedure is processed according to the quality information and the quality inspection instruction. The embodiment combines the image data on the vision and the point cloud data on the touch to construct a quality inspection framework of furniture, realizes a general furniture quality inspection scheme, decouples the processing business and the quality inspection logic of the furniture, flexibly adjusts the programs and the labels therebetween under the quality inspection framework of the furniture, can quickly follow up the quality inspection scheme when the production equipment is slightly adjusted, reduces the learning cost of the quality inspection, and the image data on the vision and the point cloud data on the touch can touch the small defects of the furniture, the accuracy of the quality inspection is high, and the situation of missed inspection is greatly reduced, thereby effectively improving the efficiency of the furniture production and improving the yield of the furniture production as a whole.

[0164] It should be noted that the size of the serial number of each step in the above embodiment does not mean the order of execution, the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0165] Referring to Figure 4 , a schematic diagram of a furniture production control device provided by an embodiment of the present application is shown, which can specifically include the following modules:

[0166] The processing equipment determination module 301 is configured to determine a plurality of processing equipment on a production line of furniture, each of the processing equipment being configured to process a part of the furniture, and each of the processing equipment being provided with a trigger, a camera and a radar.

[0167] The raw data acquisition module 302 is configured to simultaneously acquire raw image data of the part by the camera and acquire raw point cloud data of the part by the radar when the part processed by the processing equipment is transported to a track and the part activates the trigger.

[0168] The target data screening module 303 is configured to compare the raw image data and the raw point cloud data with each other to screen target image data representing the part from the raw image data and screen target point cloud data representing the part from the raw point cloud data.

[0169] The configuration file reading module 304 is configured to read a configuration file of the processing equipment, and the configuration file records one or more quality inspection labels.

[0170] The quality inspection program loading module 305 is configured to load a quality inspection program corresponding to the quality inspection label according to a similarity of the processing equipment in processing the part.

[0171] The part quality inspection module 306 is configured to start the quality inspection program, perform a quality inspection operation on the part according to the target image data and / or the target point cloud data, and obtain quality information.

[0172] The part model construction module 307 is configured to construct a three-dimensional part model of the part according to the target image data and the target point cloud data.

[0173] The quality inspection processing module 308 is configured to, if a quality inspection instruction is received when the quality information is displayed on the part model, process the part according to the quality inspection instruction, and process the quality inspection program according to the quality information and the quality inspection instruction.

[0174] In an embodiment of the present application, the target data screening module 303 comprises:

[0175] The conversion matrix reading module is configured to read a conversion matrix for calibration between the camera and the radar.

[0176] The raw point cloud data projection module is configured to project the raw point cloud data into the raw image data according to the conversion matrix to obtain reorganized image data.

[0177] Reference image data query module, configured to query reference image data collected by the camera when the machining equipment does not process the part;

[0178] First candidate image data calculation module, configured to calculate difference image data between the original image data and the reference image data as first candidate image data;

[0179] Reference surface construction module, configured to construct a reference surface representing the track according to the original point cloud data;

[0180] Second candidate image data screening module, configured to screen data above the reference surface from the reorganized image data to obtain second candidate image data;

[0181] Feature image data generation module, configured to intersect the first candidate image data and the second candidate image data to obtain feature image data;

[0182] Target data determination module, configured to determine that pixel points in the feature image data constitute target image data representing the part, and points in the feature image data constitute target point cloud data representing the part.

[0183] In an embodiment of the present application, the first candidate image data calculation module comprises:

[0184] Difference image data calculation module, configured to calculate difference image data between the original image data and each frame of the reference image data if there are multiple frames of the reference image data;

[0185] Area statistics module, configured to count the area of foreground data in the difference image data;

[0186] Area selection module, configured to select the difference image data with the smallest area of the foreground data to obtain the first candidate image data.

[0187] In an embodiment of the present application, the quality inspection procedure loading module 305 comprises:

[0188] Process information reading module, configured to read process information for processing the part set for the machining equipment;

[0189] Keyword screening module, configured to screen a plurality of keywords from the process information;

[0190] Process vector conversion module, configured to convert the plurality of keywords into a process vector of the machining equipment;

[0191] The similarity calculation module is configured to calculate a similarity between the process vector of the current processing device and the process vector of another processing device;

[0192] The candidate device determination module is configured to determine the another processing device as a candidate device if the similarity is greater than or equal to a preset first threshold value.

[0193] The candidate device loading module is configured to load the quality inspection program corresponding to the quality inspection label from the candidate device if the candidate device has configured the quality inspection program corresponding to the quality inspection label and a running time of the quality inspection program corresponding to the quality inspection label on the candidate device exceeds a preset second threshold value.

[0194] In an embodiment of the present application, the candidate device loading module comprises:

[0195] The version information query module is configured to query first version information of the quality inspection program corresponding to the quality inspection label in the current processing device and second version information of the quality inspection program corresponding to the quality inspection label in the candidate device respectively if the current processing device has configured the quality inspection program corresponding to the quality inspection label.

[0196] The differential upgrade package downloading module is configured to download a differential upgrade package between the first version information and the second version information from the candidate device if the first version information is lower than the second version information and the quality inspection program corresponding to the quality inspection label in the current processing device is not prohibited from being upgraded to the second version information.

[0197] The quality inspection program upgrading module is configured to upgrade the quality inspection program corresponding to the quality inspection label in the current processing device using the differential upgrade package.

[0198] In an embodiment of the present application, the component quality inspection module 306 comprises:

[0199] The quality inspection program starting module is configured to start the quality inspection program; the quality inspection program comprises a quality inspection network; the quality inspection network comprises a first feature structure, a second feature structure, a first header structure and a second header structure.

[0200] The gray image data conversion module is configured to convert feature image data composed of the target image data and the target point cloud data into gray image data.

[0201] The public feature extraction module is configured to input the gray image data into the first feature structure to extract public features.

[0202] The crack data segmentation module is configured to input the public features into the first header structure to segment crack data on the component.

[0203] The private domain feature conversion module is configured to convert the public domain feature into a private domain feature in the second feature structure.

[0204] The defect data detection module is configured to input the private domain feature into the second head structure to detect defect data on the part;

[0205] The abnormal index statistics module is configured to count a first abnormal index for the crack data and a second abnormal index for the defect data.

[0206] The quality inspection requirement information query module is configured to query a first quality inspection requirement information and a second quality inspection requirement information set for the part.

[0207] The quality pass determination module is configured to determine that the quality information of the part is qualified if the first abnormal index meets the first quality inspection requirement information and the second abnormal index meets the second quality inspection requirement information.

[0208] The quality unqualified determination module is configured to set the crack data and the defect data as the quality information of the part if the first abnormal index does not meet the first quality inspection requirement information and / or the second abnormal index does not meet the second quality inspection requirement information.

[0209] In an embodiment of the present application, the first feature structure includes a first convolution block and a second convolution block, the second feature structure includes a residual network, a pooling layer and a multi-layer perception, the first head structure includes a first deconvolution block and a second deconvolution block, the second head structure includes a third convolution block and a plurality of fully connected layers, and the public domain feature includes a first basic feature and a second basic feature.

[0210] The public domain feature extraction module is further configured to:

[0211] The grayscale image data is input into the first convolution block to perform a convolution operation to obtain a first basic feature.

[0212] The first basic feature is input into the second convolution block to perform a convolution operation to obtain a second basic feature.

[0213] The crack data segmentation module is further configured to:

[0214] The second basic feature is input into the first deconvolution block to perform a deconvolution operation to obtain a first reconstructed feature.

[0215] The first reconstructed feature and the first basic feature are spliced into a second reconstructed feature.

[0216] The second reconstruction feature is input into the second deconvolution block to perform a deconvolution operation to obtain crack data on the part;

[0217] The private domain feature conversion module is further configured to:

[0218] The first basic feature and the second basic feature are spliced into a third basic feature;

[0219] The third basic feature is input into the residual network to extract a first intermediate feature;

[0220] The first intermediate feature is input into the pooling layer to perform a pooling operation to obtain a second intermediate feature;

[0221] The second intermediate feature is input into the multi-layer perception to be mapped into a private domain feature;

[0222] The defect data detection module is further configured to:

[0223] The private domain feature is input into the third convolution block to perform a convolution operation to obtain a target feature;

[0224] The target feature is sequentially input into a plurality of fully connected layers to map the defect data on the part.

[0225] In an embodiment of the present application, the part model construction module 307 comprises:

[0226] A model backbone structure construction module is configured to construct a three-dimensional model backbone structure for the part using the target point cloud data;

[0227] A texture map extraction module is configured to extract a texture map from the target image data;

[0228] A texture map rendering module is configured to render the texture map into the model backbone structure to obtain a part model;

[0229] The quality inspection processing module 308 comprises:

[0230] A processing permission module is configured to, if the quality inspection instruction represents successful quality inspection, permit the part to be transported to a next processing equipment for processing;

[0231] A processing prohibition module is configured to, if the quality inspection instruction represents failed quality inspection, remove the part from the production line and prohibit the part from being transported to a next processing equipment for processing;

[0232] The quality inspection processing module 308 comprises:

[0233] The statistical index calculation module is configured to, if the quality inspection program is upgraded from the first version information to the second version information, compare the quality information with the quality inspection instruction to calculate a plurality of statistical indexes.

[0234] The effective determination module is configured to, if the plurality of statistical indexes satisfy a preset performance condition, determine that the quality inspection operation of the quality inspection program under the second version information on the processing equipment is effective.

[0235] The invalid determination module is configured to, if the plurality of statistical indexes do not satisfy the preset performance condition, determine that the quality inspection operation of the quality inspection program under the second version information on the processing equipment is invalid, and rollback the quality inspection program from the second version information to the first version information.

[0236] The furniture production control device provided by the embodiment of the present application can be applied to implement each step in each method embodiment.

[0237] For the device embodiment, it is basically similar to the method embodiment, so it is described more simply, and the related parts refer to the description in the method embodiment.

[0238] Referring to Figure 4 , a schematic diagram of a terminal device provided by an embodiment of the present application is shown. As shown in Figure 1 , the terminal device 400 in the embodiment of the present application includes a processor 410, a memory 420, and a computer program 421 stored in the memory 420 and executable on the processor 410. The processor 410 implements the steps in each embodiment of the furniture production control method when executing the computer program 421, for example Figure 3 , steps 101 to 108. Alternatively, the processor 410 implements the functions of each module / unit in each device embodiment when executing the computer program 421, for example Figure 4 , the functions of the processing equipment determination module 301, the original data acquisition module 302, the target data screening module 303, the configuration file reading module 304, the quality inspection program loading module 305, the component quality inspection module 306, the part model construction module 307, and the quality inspection processing module 308.

[0239] Illustratively, the computer program 421 can be segmented into one or more modules / units stored in the memory 420 and executed by the processor 410 to accomplish the present application. The one or more modules / units can be a series of computer program instruction segments capable of accomplishing a specific function, which can be used to describe the execution process of the computer program 421 in the terminal device 400. For example, the computer program 421 can be segmented into a processing equipment determination module, an original data acquisition module, a target data screening module, a configuration file reading module, a quality inspection program loading module, a part quality inspection module, a part model construction module, and a quality inspection processing module, and the specific functions of each module are as follows:

[0240] The processing equipment determination module is configured to determine a plurality of processing equipment on a production line of furniture; each of the processing equipment is configured to process a part of the furniture, and each of the processing equipment is configured with a trigger, a camera, and a radar.

[0241] The original data acquisition module is configured to simultaneously call the camera to acquire original image data of the part and call the radar to acquire original point cloud data of the part when the processing equipment transports the part that has completed processing to a track and the part activates the trigger.

[0242] The target data screening module is configured to compare the original image data and the original point cloud data with each other to screen target image data representing the part from the original image data and screen target point cloud data representing the part from the original point cloud data.

[0243] The configuration file reading module is configured to read a configuration file of the processing equipment; the configuration file records one or more quality inspection labels.

[0244] The quality inspection program loading module is configured to load a quality inspection program corresponding to the quality inspection label according to the similarity of the processing equipment in processing the part.

[0245] The part quality inspection module is configured to start the quality inspection program, perform a quality inspection operation on the part according to the target image data and / or the target point cloud data, and obtain quality information.

[0246] The part model construction module is configured to construct a three-dimensional part model of the part according to the target image data and the target point cloud data.

[0247] The quality inspection processing module is configured to process the part according to a quality inspection instruction if the quality inspection instruction is received when the quality information is displayed on the part model, and process the quality inspection program according to the quality information and the quality inspection instruction.

[0248] The terminal device 400 can be the furniture production control device in the foregoing various embodiments, which can be a desktop computer, a cloud server, or the like. The terminal device 400 can include, but is not limited to, a processor 410 and a memory 420. Those skilled in the art can understand that the terminal device 400 can include more or fewer components than those shown, or can combine some components, or include different components, for example, the terminal device 400 can further include an input / output device, a network access device, a bus, and the like. ​ The terminal device 400 shown in FIG. 4 is merely an example and does not constitute a limitation on the terminal device 400, and can include more or fewer components than those shown, or can combine some components, or include different components, for example, the terminal device 400 can further include an input / output device, a network access device, a bus, and the like.

[0249] The processor 410 can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0250] The memory 420 can be an internal storage unit of the terminal device 400, such as a hard disk or a memory of the terminal device 400. The memory 420 can also be an external storage device of the terminal device 400, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, or the like. Further, the memory 420 can include both an internal storage unit and an external storage device of the terminal device 400. The memory 420 is used to store the computer program 421 and other programs and data required by the terminal device 400. The memory 420 can also be used to temporarily store data that has been output or will be output.

[0251] The terminal device disclosed in the embodiments of the present application includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the furniture production control method as described in the foregoing various embodiments when executing the computer program.

[0252] The embodiment of the present application further discloses a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to realize the furniture production control method according to the foregoing embodiments.

[0253] The embodiment of the present application further discloses a computer program product, which, when running on a computer, enables the computer to execute the furniture production control method according to the foregoing embodiments.

[0254] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than limit them. Although the present application is described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can be modified, or some technical features can be replaced by equivalent ones, and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A furniture production control method, characterized by, The application relates to a quality inspection method for a furniture production line. The method comprises the following steps: determining a plurality of processing devices on the furniture production line; each processing device is used for processing a part of the furniture, and each processing device is provided with a trigger, a camera and a radar; when the processing device delivers the processed part to a track and the part activates the trigger, simultaneously calling the camera to collect original image data of the part and calling the radar to collect original point cloud data of the part; comparing the original image data and the original point cloud data to screen target image data representing the part from the original image data and screen target point cloud data representing the part from the original point cloud data; reading a configuration file of the processing device; the configuration file records one or more quality inspection labels; loading a quality inspection program corresponding to the quality inspection label according to the similarity of the processing device in processing the part; starting the quality inspection program, performing a quality inspection operation on the part according to the target image data and / or the target point cloud data to obtain quality information; constructing a three-dimensional part model of the part according to the target image data and the target point cloud data; 2. The method of claim 1, wherein, if a quality inspection instruction is received when the quality information is displayed on the part model, processing the part according to the quality inspection instruction and processing the quality inspection program according to the quality information and the quality inspection instruction. The comparison of the original image data and the original point cloud data to screen target image data representing the part from the original image data and screen target point cloud data representing the part from the original point cloud data comprises the following steps: reading a conversion matrix for the calibration between the camera and the radar; projecting the original point cloud data into the original image data according to the conversion matrix to obtain reorganized image data; inquiring reference image data collected by the camera when the processing device does not process the part; calculating difference image data between the original image data and the reference image data as first candidate image data; constructing a reference surface representing the track according to the original point cloud data; screening data above the reference surface in the reorganized image data to obtain second candidate image data; taking the intersection of the first candidate image data and the second candidate image data to obtain feature image data; 3. The method of claim 2, wherein, determining that pixel points in the feature image data form target image data representing the part, and that points in the feature image data form target point cloud data representing the part. The calculation of difference image data between the original image data and the reference image data as first candidate image data comprises the following steps: if there are multiple frames of reference image data, calculating difference image data between the original image data and each frame of reference image data; counting the area of foreground data in the difference image data; selecting the difference image data with the smallest area of foreground data to obtain first candidate image data.

4. The method of claim 1, wherein, The loading the quality inspection program corresponding to the quality inspection label according to the similarity of the processing equipment in processing the part comprises: reading process information of processing the part set to the processing equipment; screening a plurality of keywords from the process information; converting the plurality of keywords into a process vector of the processing equipment; calculating the similarity between the process vector of the current processing equipment and the process vector of other processing equipment; if the similarity is greater than or equal to a preset first threshold, determining other processing equipment as a candidate equipment; if the candidate equipment has configured the quality inspection program corresponding to the quality inspection label, and the running time of the quality inspection program corresponding to the quality inspection label on the candidate equipment exceeds a preset second threshold, loading the quality inspection program corresponding to the quality inspection label from the candidate equipment.

5. The method of claim 4, wherein, The loading the quality inspection program corresponding to the quality inspection label from the candidate equipment comprises: if the current processing equipment has configured the quality inspection program corresponding to the quality inspection label, respectively querying first version information of the quality inspection program corresponding to the quality inspection label in the current processing equipment and second version information of the quality inspection program corresponding to the quality inspection label in the candidate equipment; if the first version information is lower than the second version information, and the quality inspection program corresponding to the quality inspection label in the current processing equipment is not prohibited from being upgraded to the second version information, downloading a differential upgrade package between the first version information and the second version information from the candidate equipment; upgrading the quality inspection program corresponding to the quality inspection label in the current processing equipment using the differential upgrade package.

6. The method according to any one of claims 1-5, characterized in that, The starting the quality inspection program comprises: starting the quality inspection program; the quality inspection program comprises a quality inspection network; the quality inspection network comprises a first feature structure, a second feature structure, a first head structure and a second head structure; converting feature image data composed of the target image data and the target point cloud data into gray image data; inputting the gray image data into the first feature structure to extract public domain features; inputting the public domain features into the first head structure to segment crack data on the part; inputting the public domain features into the second feature structure to convert into private domain features; inputting the private domain features into the second head structure to detect defect data on the part; statistically calculating a first abnormal index of the crack data and a second abnormal index of the defect data; inquiring first quality inspection requirement information and second quality inspection requirement information set to the part; if the first abnormal index meets the first quality inspection requirement information, and the second abnormal index meets the second quality inspection requirement information, determining that the quality information of the part is quality qualified; if the first abnormal index does not meet the first quality inspection requirement information, and / or the second abnormal index does not meet the second quality inspection requirement information, setting the crack data, the defect data and quality unqualified as the quality information of the part.

7. The method of claim 6, wherein, The first feature structure includes a first convolutional block and a second convolutional block, the second feature structure includes a residual network, a pooling layer and a multi-layer perceptron, the first head structure includes a first deconvolutional block and a second deconvolutional block, and the second head structure includes a third convolutional block and a plurality of fully connected layers; the public domain feature includes a first basic feature and a second basic feature; The inputting the grayscale image data into the first feature structure to extract a public domain feature comprises: performing a convolution operation on the grayscale image data in the first convolutional block to obtain a first basic feature; performing a convolution operation on the first basic feature in the second convolutional block to obtain a second basic feature; The inputting the public domain feature into the first head structure to segment crack data on the part comprises: performing a deconvolution operation on the second basic feature in the first deconvolutional block to obtain a first reconstructed feature; splicing the first reconstructed feature and the first basic feature into a second reconstructed feature; performing a deconvolution operation on the second reconstructed feature in the second deconvolutional block to obtain crack data on the part; The inputting the public domain feature into the second feature structure to convert it into a private domain feature comprises: splicing the first basic feature and the second basic feature into a third basic feature; inputting the third basic feature into the residual network to extract a first intermediate feature; performing a pooling operation on the first intermediate feature in the pooling layer to obtain a second intermediate feature; mapping the second intermediate feature into a private domain feature in the multi-layer perceptron; The inputting the private domain feature into the second head structure to detect defect data on the part comprises: performing a convolution operation on the private domain feature in the third convolutional block to obtain a target feature; sequentially inputting the target feature into a plurality of fully connected layers to map defect data on the part.

8. The method according to any one of claims 1-5, characterized in that, The constructing a three-dimensional part model according to the target image data and the target point cloud data comprises: constructing a three-dimensional model skeleton structure of the part using the target point cloud data; extracting a texture map from the target image data; rendering the texture map to the model skeleton structure to obtain a part model; The processing the part according to the quality inspection instruction comprises: if the quality inspection instruction indicates successful quality inspection, allowing the part to be transported to a next processing equipment for processing; if the quality inspection instruction indicates failed quality inspection, removing the part from the production line and prohibiting the part from being transported to the next processing equipment for processing; The processing the quality inspection procedure according to the quality information and the quality inspection instruction comprises: if the quality inspection procedure is upgraded from first version information to second version information, comparing the quality information and the quality inspection instruction to calculate a plurality of statistical indicators; if a plurality of the statistical indicators satisfy a preset performance condition, determining that the quality inspection procedure under the second version information is effective for quality inspection operation of the processing equipment. If the plurality of statistical indicators do not satisfy the preset performance condition, it is determined that the quality inspection procedure under the second version information is invalid for the quality inspection operation of the processing equipment, and the quality inspection procedure is rolled back from the second version information to the first version information.

9. A terminal device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The computer program is executed by the processor to implement the furniture production control method according to any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 9. The computer program is executed by the processor to implement the furniture production control method according to any one of claims 1-8.

Images