CN-121982616-A - Abnormal dynamic detection method in feed preparation process

CN121982616ACN 121982616 ACN121982616 ACN 121982616ACN-121982616-A

Abstract

The invention discloses an abnormal dynamic detection method in a feed preparation process, in particular to the field of feed preparation monitoring, which is used for solving the problems that the abnormal state of a screening outlet is difficult to stably identify along with the time evolution and lack of rechecking evidence; through establishing a unit time throughput sequence and synchronous video time slice index relation under the same time reference, extracting a material flow motion intensity sequence and a material flow continuity sequence in a fixed observation area, constructing a dynamic base line in an inter-interaction stable segment, calculating a multi-sequence deviation track, applying a persistence constraint and a consistency constraint to obtain an abnormal evolution segment, generating an abnormal evidence chain record, and matching a preset abnormal type library to output an abnormal dynamic judgment result and a treatment clue prompt.

Inventors

ZHANG GANGJUAN
SONG YUXUAN
LIU JIANFENG
DAI MENG

Assignees

陕西杨凌瑞祺生物科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260224

Claims (10)

1. The abnormal dynamic detection method in the feed preparation process is characterized by comprising the following steps: S1, synchronously acquiring accumulated throughput and synchronous video of the same view field at a screening outlet, and calculating to obtain a throughput sequence in unit time based on the accumulated throughput and a corresponding time interval after aligning with a time stamp; S2, determining a fixed observation area for the synchronous video, and generating a stream motion intensity sequence and a stream continuity sequence according to the time granularity of the throughput sequence of unit time, wherein the stream motion intensity sequence is obtained by summarizing the motion quantity of the difference of adjacent frames, and the stream continuity sequence is obtained by summarizing the spatial continuity of the motion quantity in the observation area; S3, carrying out time sequence alignment on the throughput sequence in unit time, the stream motion intensity sequence and the stream continuity sequence, screening according to fluctuation consistency of the throughput sequence, the stream motion intensity sequence and the stream continuity sequence in unit time in a sliding time slice to obtain an inter-effect stable segment, and constructing a dynamic base line based on the inter-effect stable segment, wherein the dynamic base line simultaneously comprises the corresponding relation between the throughput in unit time, the stream motion intensity and the stream continuity; s4, calculating a multi-sequence deviation track by taking a dynamic baseline as a reference in a full period, merging intervals meeting constraint of persistence and consistency in the deviation track into abnormal evolution fragments, and outputting a start-stop boundary of each abnormal evolution fragment and a deviation form of each abnormal evolution fragment on the multi-sequence; s5, selecting a key frame for each abnormal evolution segment according to the turning position of the deviation form, intercepting the corresponding throughput segment in unit time, generating an abnormal evidence chain record containing the key frame and the corresponding segment, and outputting an abnormal dynamic judgment result and a treatment clue prompt based on matching of the multi-sequence deviation form and a preset abnormal form library in the abnormal evidence chain record for associated archiving.
2. The method for dynamically detecting anomalies in a feed preparation process according to claim 1, wherein step S1 includes: The screening outlet position is provided with a weighing and metering unit and an imaging unit which share the same time reference to generate a time stamp, the weighing and metering unit outputs accumulated throughput and converts the accumulated throughput according to the time interval of adjacent time stamps to obtain a unit time throughput sequence, and the unit time throughput sequence is divided into synchronous time slices according to the time stamps and establishes a time slice index relationship with the corresponding time slices of the synchronous video.
3. The method for dynamically detecting anomalies in a feed preparation process according to claim 2, characterized in that step S2 comprises: The synchronous video takes a frame set according to a time slice index relation, the frame set is ordered according to a frame time stamp, a frame picture is cut into a fixed observation area and gray scale normalization is completed, the fixed observation area generates a motion frequency chart by continuous frame difference and is divided by a double peak valley threshold value to obtain a motion frequency salient area, and the motion frequency salient area is determined by decomposing a four-neighborhood connected domain and selecting a maximum connected domain external rectangle.
4. A method for dynamic detection of anomalies in a feed preparation process as claimed in claim 3, characterized in that step S2 further comprises: The motion quantity distribution is summarized according to time slices to generate a material flow motion intensity sequence and shares a time slice index with a unit time throughput sequence, the motion quantity distribution is binarized by a bimodal valley threshold to obtain a motion mask, and the main connected domain duty ratio is calculated by decomposing the four-adjacent domain connected domain, so that the main connected domain duty ratio is summarized according to time slices to generate a material flow continuity sequence.
5. The method for dynamically detecting anomalies in a feed preparation process according to claim 4, wherein step S3 includes: the unit time throughput sequence, the stream motion intensity sequence and the stream continuity sequence are aligned according to time slice indexes and the time slices without frame marks are removed, the sliding time slices are constructed and divided into a first half set and a second half set on the time slice indexes, the median of the first half set and the median of the second half set are used for generating direction representation, and the maximum and minimum spans of the sliding time slices are used for generating relative amplitudes.
6. The method for dynamically detecting anomalies in a feed preparation process according to claim 5, wherein step S3 further comprises: The direction consistency indication is obtained by completely identical three direction characterizations, the amplitude similarity is obtained by calculating the minimum value and the maximum value of three relative amplitudes, the consistency degree sequence is obtained by combining the direction consistency indication and the amplitude similarity, the cross-correlation stable segment is screened according to the local peak value of the consistency degree sequence and is combined with the adjacent index to form a section, and the dynamic base line is divided into sections in the cross-correlation stable segment according to the division points and generates a partition mapping relation and an allowable fluctuation range.
7. The method for dynamically detecting anomalies in a feed preparation process according to claim 6, wherein step S4 includes: The base line reference fragment is determined according to the inter-evidence stable fragment, the base line reference fragment takes the corresponding inter-evidence stable fragment when the time slice index falls into the inter-evidence stable fragment, the base line reference fragment takes the inter-evidence stable fragment with the end time slice index closest to and earlier than the time slice index when the time slice index does not fall into the inter-evidence stable fragment, and the partition unit number is determined by the number of the stream movement intensity sequence value falling into the intensity interval and the number of the stream continuity sequence value falling into the continuity interval together.
8. The method for dynamically detecting anomalies in a feed preparation process according to claim 7, wherein step S4 further comprises: The unit time throughput deviation amount is judged according to the allowable fluctuation range corresponding to the partition unit number, the dimensionless deviation amount is formed, the stream movement intensity deviation amount and the stream continuity deviation amount are formed according to the segment-level baseline triplets, the multi-sequence deviation track is obtained by arranging the unit time throughput deviation amount, the stream movement intensity deviation amount and the stream continuity deviation amount according to time slice indexes, the continuous candidate interval is decomposed according to the deviation existence mark sequence continuous section and screened according to the continuity constraint, the consistency screening is completed according to the same direction rule and the abnormal evolution segments are obtained according to the interval condition in a merging mode, and the deviation form description is recorded.
9. The method for dynamically detecting anomalies in a feed preparation process according to claim 8, wherein step S5 includes: And reading a multi-sequence deviation track in the abnormal evolution segment, combining the deviation morphology description to generate a time point set, wherein the time point set comprises an abnormal evolution segment start time slice index and an abnormal evolution segment end time slice index, and comprises a maximum deviation intensity turning index and a dominant deviation type switching index, and the time slice index relation is used for mapping the time slice indexes in the time point set to the synchronous video frame set and reserving a frame-free mark.
10. The method for dynamically detecting anomalies in a feed preparation process according to claim 9, wherein step S5 further comprises: The time slice index in the time point set is used for extracting a key frame and intercepting a unit time throughput fragment, the key frame is determined by a frame-level motion intensity maximum corresponding frame in the synchronous video frame set and is determined by a frame time stamp close to the center of the time slice when the key frame is parallel, the abnormal evidence chain record is generated by combining the key frame, the unit time throughput fragment and a deviation morphology description and is endowed with a unique index number, and the preset abnormal pattern library is matched, and based on a deviation direction combination template and a dominant deviation type sequence template, an abnormal dynamic judgment result and a treatment clue prompt are output and associated and archived according to the unique index number.

Description

Abnormal dynamic detection method in feed preparation process Technical Field The invention relates to the field of feed preparation monitoring, in particular to an abnormal dynamic detection method in the feed preparation process. Background The feed preparation is usually produced by a continuous production line structure, and upstream ingredients and mixing, tempering and granulating can continuously bring the material state to the downstream, so that the final appearance of finished product classification and delivery forms is formed only after screening and packaging links. Taking a Chinese patent document 'an intelligent control system for feed production' (publication No. CN118321140A, application No. 202410752024.6, and Google patent retrieval website as a source) as an example, key links in feed processing are focused on classification screening and packaging quality analysis, modules such as particle feed number statistics, screening accuracy analysis, screening abnormality control and packaging quality analysis are arranged, a screening effect is evaluated by acquiring a particle image through a camera and combining weighing information, and meanwhile, weight collection and tightness related test is carried out on an extracted sample in a packaging link to form quality feedback, so that online monitoring and management of two end procedures of screening and packaging are realized. In the scheme of the comparison document, the triggering path of the abnormal control is driven by a single-procedure index, whether the screening precision is lower than a set value or not is used as a condition for entering the abnormal control flow in the screening link, the profile of the screening hole, the screening vibration frequency and the throughput in the monitoring time period are collected after entering, and the abnormal reason is confirmed and regulated by combining the normal throughput and the compensation relation stored in the database. The design is easy to encounter a contradiction which is not intuitive but truly exists in an engineering site, that a lot of anomalies in feed preparation are not jumped to screening indexes to cross the boundary at a certain moment, but accumulated in a slow drifting mode from the upstream, and are shown as continuous deviation of particle morphology, material flow rhythm, particle size distribution, pulverization trend and the like along with time, when the deviation is still early, screening accuracy can still fall near a threshold value without triggering an abnormal flow, but the anomalies are transmitted among a plurality of working procedures and gradually amplified, more end result signals are obtained by a system when the screening link is finally out of the boundary, and the situation that the anomalies start from where, how to evolve and at which working procedure critical turning occurs is difficult to restore, so that later adjustment is more than the remedy of the end of the results, rather than the early suppression of the abnormal formation process, and a dynamic evidence chain capable of rechecking and traceability is difficult to support rapid positioning and stable early warning. In order to solve the above problems, a technical solution is now provided. Disclosure of Invention In order to overcome the defects in the prior art, the embodiment of the invention provides an abnormal dynamic detection method in the feed preparation process, which is characterized in that a throughput sequence and a synchronous video time slice index relation are established in a same time reference, a material flow motion intensity sequence and a material flow continuity sequence are extracted in a fixed observation area, a dynamic base line is established in a mutual evidence stable segment, a multi-sequence deviation track is calculated, a continuous constraint and a consistency constraint are applied to obtain an abnormal evolution segment, an abnormal evidence chain record is generated, and a preset abnormal type library is matched to output an abnormal dynamic judgment result and a treatment clue prompt, so that the problems in the background technology are solved. In order to achieve the above purpose, the present invention provides the following technical solutions: S1, synchronously acquiring accumulated throughput and synchronous video of the same view field at a screening outlet, and calculating to obtain a throughput sequence in unit time based on the accumulated throughput and a corresponding time interval after aligning with a time stamp; S2, determining a fixed observation area for the synchronous video, and generating a stream motion intensity sequence and a stream continuity sequence according to the time granularity of the throughput sequence of unit time, wherein the stream motion intensity sequence is obtained by summarizing the motion quantity of the difference of adjacent frames, and the stream continuity sequence is obtained by summarizing the spa