Search

CN-121978107-A - Intelligent visual detection and closed-loop control system for beverage cup

CN121978107ACN 121978107 ACN121978107 ACN 121978107ACN-121978107-A

Abstract

The invention belongs to the technical field of industrial control systems, and relates to an intelligent visual detection and closed-loop control system for a beverage cup, which comprises a bimodal feature acquisition module, a dual feature vector set and a dual feature vector set, wherein the bimodal feature acquisition module acquires images by using first and second exposure parameters at the same moment; the system comprises a defect feature coding module, a causal node activation module, a root cause reasoning module, a control instruction generation module and a sensitivity function generation instruction, wherein the defect feature coding module generates a salt value based on a batch number and a time stamp and executes hash operation to generate a data packet containing a semantic code, the causal node activation module indexes a process causal matrix by using the semantic code to activate a root cause node, the root cause reasoning module executes Bayesian propagation calculation by taking coordinates as constraints to generate a posterior probability vector of a process parameter, and the control instruction generation module locks the root cause parameter and calls the sensitivity function generation instruction. The invention solves the problems that the defect information loss caused by single imaging affects the detection reliability, and the existing system lacks the automatic diagnosis capability of the defect cause, depends on manual investigation adjustment to cause production interruption and feedback lag.

Inventors

  • ZHU JIALIN

Assignees

  • 广东雅斯泰包装新材料有有限公司

Dates

Publication Date
20260505
Application Date
20260114

Claims (10)

  1. 1. Intelligent visual detection and closed-loop control system towards drink cup, characterized by comprising: the dual-mode feature acquisition module is used for acquiring two frames of original images respectively with a first exposure parameter and a second exposure parameter at the same physical moment, and respectively extracting features bearing different defect modes from the two frames of original images to construct a dual feature vector set, wherein the exposure time of the first exposure parameter is smaller than that of the second exposure parameter; The defect feature coding module generates a salt value based on the current production line batch number and the time stamp, and performs hash operation on the dual feature vector set by using the salt value to generate a defect feature data packet containing semantic codes and verification information; The causal node activation module is used for receiving the defect characteristic data packet and verifying the data integrity, accessing a preset process causal matrix by using a semantic code in the defect characteristic data packet as an index after the verification is passed, and activating a root cause node representing a potential fault source; The root cause reasoning module uses root cause nodes as starting points, coordinate information analyzed from the defect characteristic data packet is used as constraint conditions, bayesian belief propagation calculation is performed on the process cause and effect matrix, and a process parameter posterior probability vector which quantifies the possibility that each process parameter is a fault source is generated; and the control instruction generation module analyzes the posterior probability vector of the technological parameter to lock the root cause parameter, calls the characteristic-parameter sensitivity function to calculate the physical adjustment quantity and generates a closed-loop control instruction.
  2. 2. The smart vision inspection and closed loop control system for drinking cups of claim 1, wherein the bimodal feature acquisition module is configured to perform the following steps: extracting an edge gradient direction histogram vector representing high-frequency edge information from an original image acquired by a first exposure parameter; Extracting color moment and contrast vectors representing the low-frequency region and color information from the original image acquired by the second exposure parameters; And writing the edge gradient direction histogram vector, the color moment and the contrast vector into a predefined data structure to complete the construction of the dual feature vector set.
  3. 3. The smart vision inspection and closed-loop control system for drinking cups of claim 1, wherein the defect feature encoding module is configured to perform the following steps: the production line batch number and the time stamp are spliced, then a hash value is calculated, and a salt value is intercepted and generated; Connecting the salt value with the serialized data of the dual feature vector set, and inputting a hash function to calculate to obtain a hash abstract; Intercepting front segment bit data of the hash abstract as a row index address, and searching corresponding semantic codes in a preset defect semantic lookup table; And intercepting the back segment bit data of the hash abstract, applying bit mask operation to generate checksum, and packaging semantic codes, the checksum and the interest region coordinates of the original image into a defect characteristic data packet.
  4. 4. The smart visual inspection and closed loop control system for drinking cups of claim 1, wherein the causal node activation module is configured to perform the following steps: regenerating a local salt value based on the synchronous production line batch number and the timestamp, performing reverse search on semantic codes in the defect characteristic data packet by using the local salt value to obtain a theoretical hash abstract and calculating a local checksum; comparing the local checksum with the checksum carried in the defect characteristic data packet through a hardware gating circuit, and outputting an enabling signal only when the local checksum and the checksum are consistent; Under the triggering of an enabling signal, reading a row vector indexed by a semantic code in a process cause and effect matrix, and determining a process parameter with a probability value exceeding a preset activation threshold in the row vector as a root cause node.
  5. 5. The intelligent visual inspection and closed-loop control system for beverage cups according to claim 1, wherein the root cause inference module is configured to perform the following operation steps: Searching a corresponding feature vector approximation value in a preset reverse lookup table by using a semantic code to serve as observation evidence; screening out related process parameter subsets from a preset process parameter-space region association table by utilizing the region of interest coordinates in the defect characteristic data packet to form a limited calculation space; And in the limited computation space, iteratively updating the prior probabilities of the root cause node and the associated nodes based on the observation evidence until the probability variation satisfies the convergence condition.
  6. 6. The smart vision inspection and closed loop control system for drinking cups of claim 1, wherein the control command generation module is configured to perform the following steps: identifying an element with highest probability in the posterior probability vector of the process parameter and exceeding a preset confidence threshold, and locking the corresponding parameter as a root cause parameter; Acquiring a feature-parameter sensitivity function bound with the root cause parameter, substituting the difference value between the feature vector approximate value retrieved in the root cause reasoning module and a preset reference value into the function, and calculating to obtain a physical adjustment quantity; and encoding the physical adjustment quantity according to the communication protocol format of the target equipment, and packaging the physical adjustment quantity into a data frame containing the target address and the function code.
  7. 7. The smart vision inspection and closed-loop control system for drinking cups of claim 2, wherein extracting edge gradient direction histogram vectors characterizing high frequency edge information comprises: Converting an original image acquired by the first exposure parameters into a gray image matrix, and calculating the gradient amplitude and direction of each pixel point; dividing an image into a plurality of cells, counting gradient direction histograms, combining adjacent cells into blocks, and carrying out normalization processing on the intra-block histograms; The histogram descriptors of all blocks are concatenated to generate an edge gradient direction histogram vector.
  8. 8. The smart vision inspection and closed-loop control system for drinking cups of claim 2, wherein extracting color moment and contrast vectors characterizing low frequency regions and color information, comprises: Converting an original image acquired by the second exposure parameters into an HSV color space, and respectively calculating first-order moment, second-order moment and third-order moment of three channels of hue, saturation and brightness to form a color moment characteristic vector; calculating a local contrast average value of a gray image corresponding to the original image acquired by the second exposure parameters; And splicing the color moment characteristic vector with the local contrast average value to generate a color moment and contrast vector.
  9. 9. The smart vision inspection and closed-loop control system for drinking cups of claim 1, further comprising a matrix update module for performing the following steps: Aiming at the root cause parameters which lead to improvement, the prior probability values of the root cause parameters in the corresponding rows of the process cause and effect matrix are increased according to the preset learning rate; The prior probability values for the other process parameters in the row are scaled down.
  10. 10. The smart vision inspection and closed-loop control system for drinking cups of claim 9, wherein the matrix updating module is further configured to perform the following steps: Triggering a global attenuation mechanism when the product quantity accumulated and processed by the system reaches a preset period threshold; and applying a preset attenuation factor to all probability elements in the process cause and effect matrix to enable the values to be regressed to uniform distribution.

Description

Intelligent visual detection and closed-loop control system for beverage cup Technical Field The invention belongs to the technical field of industrial control systems, and relates to an intelligent visual detection and closed-loop control system for a beverage cup. Background In the automatic production process of pre-packaged foods such as beverage cups and the like, quality detection is a key link for guaranteeing the safety and quality consistency of products. At present, an online detection system based on machine vision is generally deployed in a production line, an industrial camera is utilized to capture a product image, defects such as film sealing folds or abnormal liquid level are identified through an algorithm, and the automatic detection mode replaces manual visual inspection to a certain extent, so that the requirement of high-speed production is met. Existing vision inspection schemes typically employ a fixed imaging parameter strategy to acquire image data. In order to clearly capture different types of defect features, specific optical conditions are often required, for example, capturing high frequency edge information at the seal film often requires a short exposure time to suppress motion blur, while detecting low contrast level lines or dark foreign objects inside the cup relies on a longer exposure time to ensure adequate luminous flux. Conventional detection systems have difficulty in simultaneously compromising the detection requirements of such differentiation, essentially making a compromise between different imaging conditions. However, a single imaging strategy may cause loss of image information of a part of defect types in an acquisition stage, thereby affecting the reliability of a detection algorithm and having a certain risk of missing detection or false alarm. In addition, existing systems are often limited to binary decisions of pass or fail, and it is difficult to automatically infer the root process cause that caused the defect based on visual defect information. When continuous inferior products appear, the upstream equipment is often checked and adjusted manually, the production is easily interrupted, and a certain hysteresis exists in the feedback loop. Disclosure of Invention In order to solve the problems, the invention provides an intelligent visual detection and closed-loop control system for a beverage cup. An intelligent visual inspection and closed-loop control system for a drinking cup, comprising: the dual-mode feature acquisition module is used for acquiring two frames of original images respectively with a first exposure parameter and a second exposure parameter at the same physical moment, and respectively extracting features bearing different defect modes from the two frames of original images to construct a dual feature vector set, wherein the exposure time of the first exposure parameter is smaller than that of the second exposure parameter; The defect feature coding module generates a salt value based on the current production line batch number and the time stamp, and performs hash operation on the dual feature vector set by using the salt value to generate a defect feature data packet containing semantic codes and verification information; The causal node activation module receives the defect characteristic data packet and verifies the data integrity, and after verification is passed, the semantic code in the packet is used as an index to access a preset process causal matrix, so that the root cause node representing the potential fault source is activated; The root cause reasoning module uses root cause nodes as starting points, coordinate information analyzed from the defect characteristic data packet is used as constraint conditions, bayesian belief propagation calculation is performed on the process cause and effect matrix, and a process parameter posterior probability vector which quantifies the possibility that each process parameter is a fault source is generated; and the control instruction generation module analyzes the posterior probability vector of the technological parameter to lock the root cause parameter, calls the characteristic-parameter sensitivity function to calculate the physical adjustment quantity and generates a closed-loop control instruction. In a further aspect of the present invention, the bimodal feature acquisition module is configured to perform the following operation steps: extracting an edge gradient direction histogram vector representing high-frequency edge information from an original image acquired by a first exposure parameter; Extracting color moment and contrast vectors representing the low-frequency region and color information from the original image acquired by the second exposure parameters; And writing the edge gradient direction histogram vector, the color moment and the contrast vector into a predefined data structure to complete the construction of the dual feature vector set. In a further aspect of t