Machine vision driven plastic masterbatch feed supply cleaning parameter intelligent regulation system
By using machine vision recognition and intelligent control systems, the problem of inconsistent parameters caused by traditional manual experience has been solved, realizing intelligent and automated control of plastic masterbatch feeding and cleaning parameters, and improving production efficiency and the ability to handle abnormal conditions.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- DONGGUAN GAOBO PLASTIC MASCH CO LTD
- Filing Date
- 2026-03-26
- Publication Date
- 2026-06-23
Smart Images
Figure CN121903183B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent manufacturing technology for plastics, specifically to a machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters. Background Technology
[0002] In the production and processing of plastic masterbatch, the feeding and cleaning processes directly impact the quality of the final product. Traditional methods for controlling the feeding and cleaning parameters of plastic masterbatch rely heavily on manual experience. Operators subjectively adjust parameters such as feeding speed, cleaning flow rate, and temperature based on observations of the masterbatch's condition and equipment operation. This method is highly susceptible to human error; differences in experience among operators can lead to inconsistent parameter settings, thus affecting the cleaning effect and feeding stability of the masterbatch.
[0003] With the development of the plastics industry, the types of masterbatches are increasing, and their characteristics, such as particle size and impurity content, vary significantly, leading to more diverse requirements for the feeding and cleaning processes. Existing automated control systems mostly use fixed parameter templates, applicable only to specific types of masterbatches or single operating conditions, making it difficult to cope with complex and changing production scenarios. When masterbatch characteristics or production conditions change, the system cannot identify and adjust parameters in a timely manner, easily resulting in incomplete cleaning, feeding blockages, and ultimately reduced production efficiency and waste of raw materials.
[0004] Meanwhile, in actual production, the material feeding and cleaning process involves the synergistic effect of multiple parameters, and there are complex internal relationships between these parameters. For example, changes in temperature may affect the solubility characteristics of the masterbatch, thereby altering the pressure and duration required for cleaning; fluctuations in the feeding rate may lead to inaccurate detection of impurity ratios, affecting the adaptability of cleaning parameters. Traditional systems often process each parameter in isolation, neglecting the interrelationships between them, making precise control difficult.
[0005] Furthermore, existing technologies lack effective correlation analysis for different material feeding and cleaning conditions. When abnormal conditions occur, it is impossible to quickly locate the root cause of the problem and correct the parameters. Operators need to spend a lot of time troubleshooting each step, which not only increases downtime but may also lead to more serious production accidents due to delayed handling. Therefore, how to use advanced technologies to achieve intelligent control of plastic masterbatch feeding and cleaning parameters has become an urgent problem to be solved in the field of plastic masterbatch processing. Summary of the Invention
[0006] The purpose of this invention is to provide a machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters to solve the problems mentioned in the background art.
[0007] To achieve the above objectives, the present invention provides a machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters, the system comprising:
[0008] The image acquisition and feature extraction module, based on the operating status of the plastic masterbatch feeding and cleaning equipment, identifies and extracts the feeding speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure and cleaning time elements in the image domain, maps the elements to preset parameter fields and establishes internal association identifiers, and establishes standardized feeding and cleaning data units.
[0009] The parameter matching and control rule generation module, based on the standardized material feeding and cleaning data unit, retrieves control rule conditions and clauses from the database that match the material feeding and cleaning type and process. It calculates the parameter control complexity factor based on the number of matching clauses and the nesting level. Based on the parameter control complexity factor, it compares the material feeding and cleaning parameters with the clause constraint values, timeliness and working condition range item by item, determines the attribute matching status, and generates a single parameter matching status determination.
[0010] The multi-condition association modeling module, based on multiple standardized material feeding and cleaning data units, finds shared entities of material feeding speed, associated temperature and pressure among different material feeding and cleaning conditions, calculates the cross-condition association density index based on the number and category of shared entities, and constructs a graph structure with working condition entities as nodes and shared relationships as edges to establish a multi-condition association graph.
[0011] The abnormal state identification and parameter correction module calculates a risk score based on the multi-condition correlation map and the matching status of individual parameters of the associated conditions, and generates a comprehensive control risk warning.
[0012] Preferably, the acquisition steps of the standardized feeding and cleaning data unit are as follows: based on the operating status of the plastic masterbatch feeding and cleaning equipment, collect equipment monitoring images one by one, perform image feature recognition and matching, and perform image domain content splitting, extracting the feeding speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure and cleaning time elements one by one, and generating the original set of operating elements of the feeding and cleaning equipment.
[0013] Based on the original set of elements for the operation of the feeding and cleaning equipment, the parameters are mapped and matched one by one, and the content of the successfully matched parameter fields is structurally converted and reconstructed to generate a standardized set of mapping fields for the operation of the feeding and cleaning equipment.
[0014] Based on the standardized mapping field set of the feeding and cleaning equipment, internal association matching is performed between each standardized mapping field, and an internal association relationship identifier is established according to the matching result. The index of the standardized mapping field is constructed and the relationship is bound to generate a standardized feeding and cleaning data unit.
[0015] Preferably, the step of obtaining the parameter control complexity factor is as follows: based on the material supply and cleaning type field and process requirement field in the standardized material supply and cleaning data unit, the original field values are extracted and matched with the field definition library for standardized field content, missing fields are filled in and the field format and semantic expression are unified to obtain a standardized set of material supply and cleaning type field and process requirement field combination;
[0016] Based on the standardized combination set of material supply cleaning type field and process requirement field, the control rule condition clauses that match the material supply cleaning type are sequentially retrieved from the database, and the clauses that match the process requirement field and combination set are filtered. The nesting level, number of logical judgments, number of times the clause is referenced, coverage of the clause activation period, overlap of applicable working conditions and frequency of call of each clause are extracted to generate a set of structured information matching control rule condition clauses.
[0017] Based on the structured information set matched with the conditions of the aforementioned control rules, the parameter control complexity factor is calculated.
[0018] Preferably, the step of obtaining the single parameter matching status determination is as follows: based on the parameter adjustment complexity factor, the material supply and cleaning type field value and the clause constraint value in the standardized material supply and cleaning data unit are called item by item, and the condition threshold comparison, field value range test and timeliness matching judgment between the field value and the constraint value are performed to form an intermediate result set of material supply and cleaning parameter and clause constraint comparison.
[0019] Based on the intermediate result set of the comparison between the material supply and cleaning parameters and the clause constraints, the working condition field value in the standardized material supply and cleaning data unit is obtained, and the working condition field of the material supply and cleaning parameters is mapped item by item to the applicable working condition range of the control rule clause. Clauses with successful working condition mapping are filtered out, and a set of matching results between the material supply and cleaning parameters and the working condition range is generated.
[0020] Based on the matching result set of the material feeding and cleaning parameters and the working condition range, the attribute matching status of each material feeding and cleaning parameter is determined, and a single parameter matching status judgment is generated.
[0021] Preferably, the step of obtaining the cross-working condition correlation tightness index is as follows: based on multiple standardized material supply and cleaning data units, extract the material supply speed field, correlation temperature field and pressure field from each standardized material supply and cleaning data unit, perform entity classification matching and deduplication processing on all material supply and cleaning data according to field type, and generate a material supply shared entity set;
[0022] Based on the set of shared entities for material supply, the occurrence count of each entity in different material supply and cleaning conditions and the category label of the entity are executed one by one. The repeated distribution information of each type of shared entity in each standardized material supply and cleaning data unit is recorded to obtain the set of shared entity distribution structure.
[0023] Based on the shared entity distribution structure set, the cross-condition correlation density index is calculated.
[0024] Preferably, the steps for obtaining the multi-condition association map are as follows: based on the cross-condition association density index, extract the feeding speed field, associated temperature field, and pressure field from multiple standardized feeding and cleaning data units, perform entity normalization identification encoding and entity type labeling on each field content, and generate a set of feeding condition entity nodes.
[0025] Calculate the sharing relationship strength value between the entities in the material supply condition based on the set of entity nodes in the material supply condition;
[0026] Based on the shared relationship strength value, a graph structure is constructed with the material supply condition entity as the node and the shared relationship strength value as the edge weight, and an entity edge connection relationship mapping table is established between the nodes to generate a multi-condition association graph.
[0027] Preferably, the step of obtaining the comprehensive control risk warning is as follows: based on the multi-condition association graph, extract the bidirectional paths of all material supply condition entity nodes from the graph structure, and retrieve the single parameter matching status judgment result, path length, path end node out-degree, path start node field number and path intermediate node number corresponding to each path to generate a multi-condition path attribute set.
[0028] Based on the multi-condition path attribute set, calculate the risk score of the corresponding entity node;
[0029] Based on the risk score, all material supply condition entity nodes are divided into risk level ranges according to the risk score, and risk level color codes and risk warning node numbers are marked in each material supply path to generate comprehensive control risk prompts.
[0030] Preferably, the generation step of the structured information set matching the control rule condition clauses is as follows: based on the standardized combination set of the material supply cleaning type field and the process requirement field, traverse all control rule condition clauses related to the material supply cleaning type in the database, extract the logical operators, numerical thresholds, time windows and working condition restrictions contained in the clauses, classify and label them according to the nesting level of the clauses, record the trigger frequency and matching accuracy of each clause in historical control, and generate a structured information set containing clause attributes and historical performance.
[0031] Preferably, the risk score calculation steps are as follows: based on the single parameter matching status judgment result in the multi-condition path attribute set, assign a basic score to each path; set weight coefficients for weighted calculation according to path length, out-degree of the path endpoint node, number of fields of the path starting node and number of intermediate nodes; after accumulating the scores of each path, make corrections in combination with the risk data of the same condition in historical control, and generate the risk score of the corresponding entity node.
[0032] Preferably, the specific steps of the image domain content splitting are as follows: the collected equipment monitoring images are divided into continuous frames according to the time sequence, each frame image is segmented into regions, the image content of the feed port region, the cleaning tank region and the temperature and pressure sensor region are extracted respectively, the independent feature recognition of each region image is performed, and a set of original elements of the feed cleaning equipment operation containing multi-region features is generated.
[0033] Compared with the prior art, the beneficial effects of the present invention are:
[0034] This system uses machine vision to intelligently control the parameters of plastic masterbatch feeding and cleaning, changing the traditional parameter adjustment mode that relies on manual experience. The image acquisition and feature extraction module can accurately identify multiple elements in the feeding and cleaning process and establish standardized data units, so that the originally scattered parameter information is systematically integrated, reducing the control errors caused by parameter confusion.
[0035] The parameter matching and control rule generation module retrieves data from a database based on standardized data units. By calculating the parameter control complexity factor and comparing each parameter with the constraint clauses, it generates a single-item parameter matching status judgment, making parameter control more targeted. It no longer simply applies a fixed template, but adjusts the settings according to the actual working conditions and the matching of rule clauses, making the parameter settings more suitable for the specific material feeding and cleaning type and process requirements.
[0036] The multi-condition correlation modeling module constructs a multi-condition correlation graph that clearly presents the shared entities and relationships between different conditions. This allows operators to intuitively understand the inherent connections between various conditions, breaking the limitations of traditional single-condition control and enabling cross-condition parameter collaborative adjustment, which helps to cope with complex and ever-changing production scenarios.
[0037] The abnormal state identification and parameter correction module combines multi-condition correlation maps and single-parameter matching status judgment to calculate risk scores. The generated comprehensive control risk alerts can promptly detect abnormalities in the material feeding and cleaning process. Compared with traditional manual inspection methods, this significantly shortens the time for abnormal identification, allowing operators to quickly take measures to correct parameters and reduce production losses and quality problems caused by persistent abnormal states.
[0038] The entire system automates and intelligently controls the feeding and cleaning parameters of plastic masterbatch, reducing reliance on manual operation experience, improving the consistency and accuracy of parameter control, and enhancing adaptability to different working conditions and efficiency in handling abnormal states. It demonstrates significant application value in the plastic masterbatch processing process. Attached Figure Description
[0039] Figure 1 This is a schematic diagram illustrating the working principle of the machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters described in this invention.
[0040] Figure 2 A flowchart for generating complexity factors for parameter adjustment;
[0041] Figure 3 A flowchart for generating the cross-condition correlation density index;
[0042] Figure 4 A flowchart for generating a structured information set for matching regulatory rule conditions and clauses;
[0043] Figure 5 A flowchart for splitting the content of an image domain. Detailed Implementation
[0044] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0045] Please see Figures 1-5 This invention provides a machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters. The system includes: an image acquisition and feature extraction module, a parameter matching and control rule generation module, a multi-condition correlation modeling module, and an abnormal state identification and parameter correction module. The specific implementation steps are as follows:
[0046] The image acquisition and feature extraction module, based on the operating status of the plastic masterbatch feeding and cleaning equipment, identifies and extracts key elements from the image domain, including feeding speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure, and cleaning time. These elements are mapped to preset parameter fields and internal association identifiers are established, creating a standardized feeding and cleaning data unit. The parameter matching and control rule generation module, based on the standardized feeding and cleaning data unit, retrieves control rule conditions from the database that match the feeding and cleaning type and process. It calculates the parameter control complexity factor based on the number and nesting level of the matching conditions, and then compares each feeding and cleaning parameter with the condition constraints based on this factor. The system assesses the value, timeliness, and operating condition range to determine attribute matching status and generate individual parameter matching status judgments. The multi-operating condition association modeling module, based on multiple standardized material feeding and cleaning data units, searches for shared entities related to material feeding speed, associated temperature, and pressure across different material feeding and cleaning operating conditions. It calculates the cross-operating condition association density index based on the number and category of shared entities and constructs a graph structure with operating condition entities as nodes and shared relationships as edges to establish a multi-operating condition association graph. The abnormal state identification and parameter correction module, based on the multi-operating condition association graph and combined with the individual parameter matching status judgments of associated operating conditions, calculates risk scores and generates comprehensive control risk warnings.
[0047] Example 1: The acquisition of standardized material feeding and cleaning data units is based on the operating status of the plastic masterbatch feeding and cleaning equipment. The specific process is as follows:
[0048] Image acquisition and raw element extraction are performed to monitor the equipment's operating status. The image acquisition device is activated to monitor and capture continuous images of the plastic masterbatch feeding and cleaning equipment in real time. The acquired monitoring images are then split into consecutive image frames according to time sequence, ensuring that each frame fully represents the equipment's operating status at the corresponding time point. Each image frame undergoes region segmentation processing, dividing the image into a feeding port area, a cleaning tank area, and a temperature and pressure sensor area using image segmentation technology. The feeding port area encompasses the channel through which the masterbatch enters the equipment and surrounding components; the cleaning tank area includes the tank body used for cleaning the masterbatch and the cleaning medium flowing inside; and the temperature and pressure sensor area focuses on the display sections of various temperature and pressure measuring instruments installed on the equipment.
[0049] Independent feature recognition is performed on each segmented image region. In the feed inlet region, image recognition technology is used to analyze the flow rate and frequency of the masterbatch to extract the feed speed element. In the cleaning tank region, the flow state and flow rate of the cleaning medium are identified to extract the cleaning flow rate element. Simultaneously, the size distribution of the masterbatch particles is analyzed to extract the masterbatch particle size element, and the proportion of impurities mixed in the masterbatch is identified to extract the impurity ratio element. In the temperature and pressure sensor region, the numerical information displayed on the instruments is identified to extract the temperature and pressure elements. Furthermore, considering the time span of image acquisition, the duration from the masterbatch entering the cleaning stage to the completion of cleaning is calculated to extract the cleaning time element. The extracted feed speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure, and cleaning time elements are integrated to form the original set of operating elements for the feed cleaning equipment.
[0050] Parameter mapping matching and verification, and standardized mapping field generation are performed. Based on the original set of operational elements of the material feeding and cleaning equipment, a preset parameter field mapping table is invoked. This table contains the correspondence between each original element and the preset parameter fields within the system. Each original element is matched and verified against the parameter fields in the mapping table, checking whether the name and meaning of the original element are consistent with the parameter field. For successfully matched parameter fields, their content structure and format are transformed. For example, the unit of temperature element is converted from Celsius to the system-standardized Kelvin, and the unit of material feeding speed is converted from meters per minute to the system-set centimeters per second. Simultaneously, field reconstruction is performed, adjusting the field arrangement order and data storage format to ensure consistency in the field structure of different elements. After transformation and reconstruction, a standardized mapping field set for the operation of the material feeding and cleaning equipment is generated.
[0051] Establish internal association identifiers and generate standardized data units. For each field in the standardized mapping field set for the operation of the material feeding and cleaning equipment, analyze the inherent relationships between the fields. For example, there is a linkage between feeding speed and cleaning flow rate; when the feeding speed increases, the cleaning flow rate usually needs to be adjusted accordingly. There is also a certain correlation between temperature and pressure; temperature changes may cause pressure fluctuations. Based on these inherent relationships, perform association matching on each standardized mapping field to determine the association type and strength between fields. Based on the matching results, add an internal association identifier to each field, including information such as the name of the associated field and the association type. Index the standardized mapping fields after adding association identifiers, establish an index table for fast querying of each field, and bind the related fields to form an organic whole.
[0052] Example 2: The process of obtaining the parameter control complexity factor and the determination of the matching status of individual parameters starts from the standardized material feeding and cleaning data unit and proceeds step by step.
[0053] The raw values of the material supply cleaning type field and process requirement field are extracted from the standardized material supply cleaning data unit, and these raw values are compared with a pre-defined field definition library. The field definition library contains standard descriptions, field formats, and semantic explanations for various material supply cleaning types and process requirements. During the comparison process, if any raw field values are missing, they are supplemented according to the association rules and historical data in the field definition library; if the format of the raw field values does not match the definition library, they are converted according to the format standards in the definition library; if there are inconsistencies in semantic expression, they are adjusted with reference to the standardized semantics in the definition library, ultimately forming a standardized set of material supply cleaning type field and process requirement field combinations.
[0054] Based on the aforementioned combination set, all control rule condition clauses related to the material supply and cleaning type are retrieved from the database. Each clause is traversed, and logical operators such as greater than, less than, equal to, AND, and OR are extracted to clarify the logical relationships between conditions within the clause. Numerical thresholds are extracted, including the specific numerical ranges of parameters such as material supply speed, cleaning flow rate, temperature, and pressure. Time windows are extracted to determine the applicable time intervals for the clauses. Operating condition restrictions are extracted to clarify which specific material supply and cleaning conditions the clauses apply to. The extracted information is categorized and labeled according to the nesting level of the clauses, such as first-level nested clauses, second-level nested clauses, etc., and the number of times each clause is triggered and successfully matched during historical control processes is recorded. This forms a structured information set of control rule condition clause matching that includes clause attributes and historical performance.
[0055] Based on the control rules and conditions, a structured information set is matched, and the number of matched clauses is counted. Different weights are assigned to clauses according to their nesting level, with higher levels having greater weights. The number of clauses and the weights at each level are combined to calculate a parameter control complexity factor, which reflects the complexity of the clauses involved in the material feeding and cleaning parameter control process.
[0056] Based on the parameter control complexity factor, feeding and cleaning parameters, such as feeding speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure, and cleaning time, are retrieved from the standardized feeding and cleaning data unit and compared item by item with the constraint values in the control rule conditions. When comparing condition thresholds, it is checked whether the feeding and cleaning parameters are within the threshold range specified in the clause; when checking the numerical range, it is confirmed whether the parameter values meet the range requirements set in the clause; when judging timeliness, it is verified whether the time corresponding to the parameter is within the applicable time window of the clause. The results of these comparisons, checks, and judgments are integrated to form an intermediate result set of the comparison between feeding and cleaning parameters and clause constraints.
[0057] The operating condition field values are extracted from the standardized material feeding and cleaning data units. These values contain information such as equipment model, masterbatch type, and production environment during the material feeding and cleaning process. The operating condition field values are mapped item by item to the applicable operating condition ranges of the control rule clauses, checking whether each piece of information in the operating condition field values matches the applicable operating condition range of the clauses. All clauses with successfully matched operating condition mappings are selected. Based on the comparison results of these clauses with the material feeding and cleaning parameters, a set of matching results between the material feeding and cleaning parameters and the operating condition ranges is generated.
[0058] Based on the matching results set of material supply and cleaning parameters and operating conditions, the matching status of each material supply and cleaning parameter with the constraint values, time limits, and operating conditions of the corresponding clauses is analyzed. For each parameter, if it matches the clause in terms of constraint values, time limits, and operating conditions, it is determined to be an attribute match; if any one of them does not match, it is determined to be an attribute mismatch, thereby generating a single parameter matching status determination.
[0059] Example 3: The process of obtaining the cross-condition correlation tightness index and multi-condition correlation map starts from multiple standardized material feeding and cleaning data units. The specific operation is as follows:
[0060] The feeding speed, associated temperature, and pressure fields are extracted from each standardized feeding and cleaning data unit. The feeding speed field contains information on the amount of plastic masterbatch passing through the feeding port per unit time. The associated temperature field covers the real-time temperature and trend data of the medium in the cleaning tank. The pressure field records the pressure values and fluctuations at key locations inside the equipment. All extracted fields are categorized by type: feeding speed, associated temperature, and pressure fields are grouped into one category. Entity matching is performed on the data in each category to identify entities with the same description or equivalent meaning. For example, fields with a feeding speed of "50 kg / h" in different data units can be considered the same entity. The matched entities are then deduplicated to remove duplicate entities, ultimately forming a shared feeding entity set. This set contains all unique entities related to feeding speed, associated temperature, and pressure that appear in different data units.
[0061] For each entity in the shared material supply entity set, the frequency of its occurrence in different material supply and cleaning conditions is counted. These conditions include different types of masterbatch, production batches, and equipment operating modes. Each entity is categorized, specifying whether it belongs to the material supply speed, associated temperature, or pressure category. Simultaneously, the distribution of each type of shared entity in each standardized material supply and cleaning data unit is recorded, including the frequency of the entity's appearance in that data unit and its coexistence relationship with other entities, thus forming a shared entity distribution structure set. Based on the data in this set, a cross-condition correlation density index is calculated. The calculation process comprehensively considers the number of shared entities, their frequency of occurrence in different conditions, and category distribution; the more conditions an entity appears in and the more evenly distributed its distribution, the higher the index value.
[0062] Based on the cross-condition correlation density index, the feeding speed, associated temperature, and pressure fields are extracted again from multiple standardized feeding and cleaning data units. Each field content is then coded using entity normalization, assigning a unique code to each unique entity to ensure that the same entity in different data units has the same code. Simultaneously, entities are type-labeled, clearly indicating whether they are feeding speed, associated temperature, or pressure entities, thereby generating a set of feeding condition entity nodes, with each node corresponding to a coded and labeled entity.
[0063] Calculate the sharing relationship strength value between entities in the material feeding condition. Analyze the coexistence of different entities in each standardized material feeding and cleaning data unit. If two entities appear simultaneously in multiple data units, it indicates that there is a sharing relationship between them. The more times they coexist, the stronger the sharing relationship value. Adjust the strength value according to the cross-condition association tightness index. The higher the index, the stronger the relationship between entities in that association.
[0064] A graph structure is constructed based on the set of entity nodes for the material supply condition and the strength values of the shared relationships between entities. In the graph structure, each entity for the material supply condition is a node, and the shared relationships between entities are represented as edges connecting nodes, with the weight of the edges determined by the strength value of the shared relationships. A mapping table of entity edge connections between nodes is established, which records the connection status of each node with other nodes, including the encoding of connected nodes and the weight of the edges.
[0065] Example 4: The acquisition of comprehensive control risk warnings begins with a multi-condition correlation graph. The specific process is as follows: From the graph structure of the multi-condition correlation graph, the system automatically traverses all possible paths and extracts bidirectional paths between the entity nodes of the material supply condition. These paths cover all possible channels from any entity node to other entity nodes, including both forward and reverse connections. For each extracted path, the system retrieves the corresponding single-parameter matching status judgment result, which shows the matching status of various parameters involved in the path with the control rule clauses. Simultaneously, it records the path length (i.e., the number of entity nodes in the path); the out-degree of the path's endpoint node (i.e., the number of connections established between the endpoint node and other nodes); the number of fields at the path's starting node (i.e., the number of material supply and cleaning parameter fields contained in the starting node); and the number of intermediate nodes traversed by the path (i.e., the total number of nodes in the path excluding the starting and ending points). This information is summarized and integrated to form a multi-condition path attribute set, which fully presents the characteristics of each path.
[0066] Based on the matching status of individual parameters in the multi-condition path attribute set, a base score is assigned to each path. If all individual parameters in the path are matched, the base score is set to a fixed initial value; if there are mismatched parameters, the initial value is reduced accordingly based on the number and importance of the mismatched parameters. The more mismatched parameters there are and the higher their importance, the greater the reduction in score.
[0067] Weighting coefficients are set based on path length, out-degree of the path's endpoint node, number of fields in the path's starting node, and number of intermediate nodes. The weighting coefficient for path length is inversely proportional to the number of nodes in the path; the more nodes, the smaller the weighting coefficient. The weighting coefficient for the out-degree of the path's endpoint node is directly proportional to the out-degree value; the larger the out-degree, the larger the weighting coefficient. The weighting coefficient for the number of fields in the path's starting node is directly proportional to the number of fields; the more fields, the larger the weighting coefficient. The weighting coefficient for the number of intermediate nodes is inversely proportional to the number of intermediate nodes; the more intermediate nodes, the smaller the weighting coefficient.
[0068] The weighted score for each path is calculated using the following formula:
[0069]
[0070] in, This represents the weighted score of the path. The basic score representing the path. Weighting coefficients representing path length. The normalized value representing the path length. The weight coefficient represents the out-degree of the endpoint node of the path. This represents the normalized value of the out-degree of the endpoint node on the path. The weighting coefficient represents the number of fields in the starting node of the path. The normalized value representing the number of fields in the path's starting node. The weighting coefficient represents the number of intermediate nodes. This represents the normalized value indicating the number of intermediate nodes. The normalized value is obtained by converting the actual value of the corresponding parameter to the range of 0-1, ensuring that the values of different parameters can be calculated on the same dimension.
[0071] The weighted scores of each path are summed to obtain a preliminary risk score. This preliminary score is then revised by incorporating risk data from historical control measures under similar operating conditions. If historical data shows that the actual risk under similar operating conditions is higher than the risk level reflected in the preliminary score, the score is appropriately increased; if the actual risk is lower than the risk level reflected in the preliminary score, the score is appropriately decreased, ultimately generating a risk score for the corresponding entity node.
[0072] Based on risk scoring, all material supply nodes are divided into different risk level ranges. These ranges typically include low, medium, and high risk, each corresponding to a specific scoring range. Nodes in the low-risk range are marked in green; those in the medium-risk range are marked in yellow; and those in the high-risk range are marked in red. Within each material supply path, nodes are labeled with corresponding color codes based on their risk level, along with the risk warning node numbers (nodes in the medium- to high-risk ranges). This information is integrated to form a comprehensive risk control alert, clearly displaying the risk nodes, risk levels, and specific locations within each material supply path.
[0073] Example 5: The specific process of image domain content segmentation and risk score calculation revolves around the operation data processing of the plastic masterbatch feeding and cleaning equipment.
[0074] Image domain content segmentation begins with the acquisition of equipment monitoring images. Image acquisition devices installed at different locations on the equipment are activated to continuously capture images of key components such as the feed inlet, cleaning tank, and temperature and pressure sensors, obtaining monitoring images covering the entire equipment operation process. These images are then segmented into consecutive image frames according to time sequence, with each frame corresponding to the equipment status at a specific point in time. Each image frame is then segmented into regions. Using a preset image segmentation algorithm, the boundaries of the feed inlet region, cleaning tank region, and temperature and pressure sensor region are defined based on the equipment's structural characteristics. The feed inlet region includes the channel through which the raw material enters the equipment and the surrounding transmission components; the cleaning tank region encompasses the raw material and cleaning medium inside the tank; and the temperature and pressure sensor region focuses on the display screens or pointer positions of various instruments.
[0075] Independent feature recognition is performed on each segmented image region. In the feed inlet region, the feed velocity is extracted by analyzing the movement trajectory and density changes of the masterbatch particles in the image. In the cleaning tank region, the cleaning flow rate is extracted based on the flow pattern and color changes of the cleaning medium; the masterbatch particle size is extracted by identifying the pixel size distribution of the masterbatch particles; and the impurity ratio is extracted by distinguishing the color and shape differences between the masterbatch and impurities and calculating the proportion of impurities in the image. In the temperature and pressure sensor region, temperature and pressure are obtained by identifying the relative position of the instrument scale and pointer or numbers. Simultaneously, the time interval from the masterbatch entering the cleaning tank to its departure is calculated by combining the timestamps of the image frames, extracting the cleaning duration. These elements are integrated to form a set of original operational elements for the feed and cleaning equipment that includes features from multiple regions.
[0076] Risk scoring is based on the matching status of individual parameters in a multi-condition path attribute set. Matching status information for each path is extracted from this set, including the matching status of each parameter with the control rule clauses. A base score is assigned to each path. When all parameters in a path match, the base score is a fixed value; if there are mismatched parameters, the score is adjusted according to the type and number of parameters. For example, if the feeding speed or temperature is mismatched, the deducted score is higher than if the cleaning time is mismatched.
[0077] The score is adjusted by combining path length, the out-degree of the path's endpoint node, the number of fields at the path's starting node, and the number of intermediate nodes. Path length is reflected by the number of entity nodes contained in the path; the more nodes, the smaller the impact on the base score. The out-degree of the path's endpoint node reflects the number of connections between that node and other nodes; the more connections, the greater the score bonus. The number of fields at the path's starting node covers the number of parameters such as feeding speed and temperature; the more fields, the greater the impact on the score. The more intermediate nodes, the more significant the weakening effect on the score.
[0078] The initial scores are adjusted based on risk data from similar operating conditions in historical control operations. If historical data shows that the same parameter combination has previously caused risk under a certain type of operating condition, the score for the corresponding path is increased accordingly; if historical data shows that the parameter combination operates stably, the score is appropriately decreased.
[0079] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0080] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters, characterized in that, The system includes: The image acquisition and feature extraction module, based on the operating status of the plastic masterbatch feeding and cleaning equipment, identifies and extracts the feeding speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure and cleaning time elements in the image domain, maps the elements to preset parameter fields and establishes internal association identifiers, and establishes standardized feeding and cleaning data units. The parameter matching and control rule generation module, based on the standardized material feeding and cleaning data unit, retrieves control rule conditions and clauses from the database that match the material feeding and cleaning type and process. It calculates the parameter control complexity factor based on the number of matching clauses and the nesting level. Based on the parameter control complexity factor, it compares the material feeding and cleaning parameters with the clause constraint values, timeliness and working condition range item by item, determines the attribute matching status, and generates a single parameter matching status determination. The multi-condition association modeling module, based on multiple standardized material feeding and cleaning data units, finds shared entities of material feeding speed, associated temperature and pressure among different material feeding and cleaning conditions, calculates the cross-condition association density index based on the number and category of shared entities, and constructs a graph structure with working condition entities as nodes and shared relationships as edges to establish a multi-condition association graph. The abnormal state identification and parameter correction module calculates a risk score based on the multi-condition correlation map and the matching status of individual parameters of the correlation conditions, and generates a comprehensive control risk warning. The steps for obtaining the parameter control complexity factor are as follows: Based on the material supply and cleaning type field and process requirement field in the standardized material supply and cleaning data unit, extract the original field values and perform standardized matching of field content with the field definition library, fill in the missing fields and unify the field format and semantic expression, and obtain a standardized set of material supply and cleaning type field and process requirement field combination. Based on the standardized combination set of material supply cleaning type field and process requirement field, the control rule condition clauses that match the material supply cleaning type are sequentially retrieved from the database, and the clauses that match the process requirement field and combination set are filtered. The nesting level, number of logical judgments, number of times the clause is referenced, coverage of the clause activation period, overlap of applicable working conditions and frequency of call of each clause are extracted to generate a set of structured information matching control rule condition clauses. Based on the structured information set matched with the aforementioned control rule conditions, the parameter control complexity factor is calculated. The steps for obtaining the cross-condition correlation tightness index are as follows: Based on multiple standardized material supply and cleaning data units, extract the material supply speed field, correlation temperature field, and pressure field from each standardized material supply and cleaning data unit, perform entity classification matching and deduplication processing on all material supply and cleaning data according to field type, and generate a set of shared material supply entities. Based on the set of shared entities for material supply, the occurrence count of each entity in different material supply and cleaning conditions and the category label of the entity are executed one by one. The repeated distribution information of each type of shared entity in each standardized material supply and cleaning data unit is recorded to obtain the set of shared entity distribution structure. Based on the shared entity distribution structure set, the cross-condition correlation density index is calculated.
2. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 1, characterized in that, The steps for acquiring the standardized feeding and cleaning data unit are as follows: based on the operating status of the plastic masterbatch feeding and cleaning equipment, collect equipment monitoring images item by item, perform image feature recognition and matching, and split the image domain content. Extract the feeding speed, cleaning flow rate, masterbatch particle size, impurity ratio, temperature, pressure and cleaning time elements one by one to generate the original set of operating elements of the feeding and cleaning equipment. Based on the original set of elements for the operation of the feeding and cleaning equipment, the parameters are mapped and matched one by one, and the content of the successfully matched parameter fields is structurally converted and reconstructed to generate a standardized set of mapping fields for the operation of the feeding and cleaning equipment. Based on the standardized mapping field set of the feeding and cleaning equipment, internal association matching is performed between each standardized mapping field, and an internal association relationship identifier is established according to the matching result. The index of the standardized mapping field is constructed and the relationship is bound to generate a standardized feeding and cleaning data unit.
3. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 1, characterized in that, The steps for obtaining the single parameter matching status determination are as follows: based on the parameter adjustment complexity factor, the material supply and cleaning type field value and the clause constraint value in the standardized material supply and cleaning data unit are called one by one, and the condition threshold comparison, field value range test and timeliness matching judgment between the field value and the constraint value are performed to form an intermediate result set of material supply and cleaning parameter and clause constraint comparison. Based on the intermediate result set of the comparison between the material supply and cleaning parameters and the clause constraints, the working condition field value in the standardized material supply and cleaning data unit is obtained, and the working condition field of the material supply and cleaning parameters is mapped item by item to the applicable working condition range of the control rule clause. Clauses with successful working condition mapping are filtered out, and a set of matching results between the material supply and cleaning parameters and the working condition range is generated. Based on the matching result set of the material feeding and cleaning parameters and the working condition range, the attribute matching status of each material feeding and cleaning parameter is determined, and a single parameter matching status judgment is generated.
4. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 1, characterized in that, The steps for obtaining the multi-condition association map are as follows: Based on the cross-condition association density index, extract the feeding speed field, associated temperature field, and pressure field from multiple standardized feeding and cleaning data units, perform entity normalization identification encoding and entity type labeling on each field content, and generate a set of feeding condition entity nodes. Calculate the sharing relationship strength value between the entities in the material supply condition based on the set of entity nodes in the material supply condition; Based on the shared relationship strength value, a graph structure is constructed with the material supply condition entity as the node and the shared relationship strength value as the edge weight, and an entity edge connection relationship mapping table is established between the nodes to generate a multi-condition association graph.
5. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 1, characterized in that, The steps for obtaining the comprehensive control risk warning are as follows: Based on the multi-condition association graph, extract the bidirectional paths of all material supply condition entity nodes from the graph structure, and retrieve the single parameter matching status judgment result, path length, path end node out-degree, path start node field number and path intermediate node number corresponding to each path to generate a multi-condition path attribute set. Based on the multi-condition path attribute set, calculate the risk score of the corresponding entity node; Based on the risk score, all material supply condition entity nodes are divided into risk level ranges according to the risk score, and risk level color codes and risk warning node numbers are marked in each material supply path to generate comprehensive control risk prompts.
6. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 1, characterized in that, The steps for generating the structured information set matching the control rule conditions are as follows: Based on the standardized combination set of material supply cleaning type field and process requirement field, traverse all control rule conditions related to material supply cleaning type in the database, extract the logical operators, numerical thresholds, time windows and operating condition restrictions contained in the conditions, classify and label them according to the nesting level of the conditions, record the trigger frequency and matching accuracy of each condition in historical control, and generate a structured information set containing condition attributes and historical performance.
7. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 6, characterized in that, The risk score calculation steps are as follows: based on the single parameter matching status judgment result in the multi-condition path attribute set, assign a basic score to each path; and perform weighted calculation by setting weight coefficients according to the path length, the out-degree of the path end node, the number of fields of the path start node, and the number of intermediate nodes. After accumulating the scores of each path, the risk scores are adjusted by combining the risk data of the same working conditions in historical regulation, and a risk score for the corresponding entity node is generated.
8. The machine vision-driven intelligent control system for plastic masterbatch feeding and cleaning parameters according to claim 2, characterized in that, The specific steps for splitting the image domain content are as follows: the collected equipment monitoring images are divided into continuous frames according to the time sequence, each frame image is segmented into regions, and the image content of the feed port region, the cleaning tank region, and the temperature and pressure sensor region are extracted respectively. Independent feature recognition is performed on the images of each region to generate a set of original elements of the feed cleaning equipment operation containing features of multiple regions.