CN-122021579-A - AI-fused intelligent manufacturing workshop instruction processing output method and system

CN122021579ACN 122021579 ACN122021579 ACN 122021579ACN-122021579-A

Abstract

The invention relates to the field of data processing, in particular to an AI-fused intelligent manufacturing workshop instruction processing output method and system, wherein the method comprises the steps of acquiring a history sample library accumulated in a history manufacturing stage, and acquiring a current recognition result obtained by ASR (automatic repeat request) conversion of worker voices in a current manufacturing stage; based on the historical sample library and the current recognition result, calculating the definition of each attribute in each form type, utilizing the definition to match a second event form corresponding to the current recognition result, dividing the current recognition result based on the matched second event form to obtain an optimal segment of an independent filling form, filling the obtained optimal segment into the second event form and outputting the optimal segment. The invention effectively suppresses the interference of matching of non-attribute words in the spoken language instruction to the table types, and avoids the information transmission abnormality caused by transmitting the instruction information to the error processing flow due to the matching error.

Inventors

YUAN PENG
XU HAO
ZHU XINHUA
ZHANG LIJIE
Wu Tongcong
LI PENGHUI

Assignees

山东青鸟工业互联网有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

An ai-fused intelligent manufacturing shop instruction processing output method is characterized by comprising the following steps: Acquiring a history sample library which is accumulated in a history manufacturing stage and comprises a form type, a professional word library, a first event form recorded as an original spoken language instruction form, a recognition result and a second event form recorded as a professional term, and acquiring a current recognition result obtained by ASR (automatic repeat request) conversion of the worker voice in the current manufacturing stage; Calculating the definition of each attribute in each form type based on the historical sample library and the current recognition result, matching the definition to obtain a second event form corresponding to the current recognition result, and dividing the current recognition result based on the matched second event form to obtain an optimal segment of the independent form filling; filling the obtained optimal fragment into a second event table and outputting the optimal fragment; The definition is comprehensively calculated based on the difference between the white words in the first event table, the difference between the proprietary terms in the second event table and the semantic proximity between the white words and the proprietary terms under each attribute; The step of matching the second event table corresponding to the current identification result by using the definition comprises the following steps: Carrying out semantic clustering on the current recognition result, respectively calculating the association strength between the cluster and the attribute from the history data association dimension and the attribute definition association dimension between each cluster and each attribute, combining the definition of each attribute, comprehensively calculating the event relativity between the current recognition result and each table type, and selecting the table type with the largest event relativity as a best-matched second event table.
2. The AI-converged intelligent manufacturing plant instruction processing output method of claim 1, wherein the calculation of the definition includes: For any attribute in any form type, acquiring all white words corresponding to the attribute in a first event form in a history sample library, acquiring all special terms corresponding to the attribute in a second event form in the history sample library, and respectively calculating a first characteristic value, a second characteristic value and a standard degree of the corresponding attribute based on the white words and the special terms; And calculating the difference value between the second characteristic value and the first characteristic value, converting the difference value between the second characteristic value and the first characteristic value into an index value through an index function, and multiplying the index value by a standard degree to obtain the product which is used as the definition of the attribute.
3. The AI-converged intelligent manufacturing plant instruction processing output method of claim 2, wherein the calculation of the first characteristic value includes: And obtaining word vectors of each white word, calculating cosine similarity between any two word vectors, calculating negative index values of the cosine similarity, and taking the average value of all the negative index values as a first characteristic value.
4. The AI-converged intelligent manufacturing plant instruction processing output method of claim 3, wherein the calculation of the second characteristic value includes: and obtaining word vectors of all the technical terms, and calculating to obtain a second characteristic value by adopting a calculation mode of the first characteristic value.
5. The AI-converged intelligent manufacturing plant instruction processing output method of claim 4, wherein the calculation of the standard degree includes: and calculating the average value of cosine similarity between each white word and the word vector of the special term at the corresponding position, and taking the average value as the standard degree.
6. The AI-converged intelligent manufacturing plant instruction processing output method of claim 1, wherein dividing the current recognition result based on the matched second event table to obtain the optimal segment of the independent form filling comprises: Word segmentation is carried out on the current recognition result to obtain a word sequence, random segmentation is carried out on the word sequence, multiple segmentation modes are generated, each segmentation mode divides the word sequence into a plurality of continuous segments, and the shortest length of each segment is not lower than the attribute number of the second event table which is matched most; for any segment in any segmentation mode, calculating the independent score of each word in the segment, and calculating the average value of the independent scores of all words in the segment; Calculating the accumulated value of Euclidean distances of similarity sequences between every two words in the segment as the dividing score of the segment; Adding the average value of the independent scores in the segments with the division scores, and averaging all the segments in the segmentation mode to obtain the comprehensive score of the segmentation mode; Traversing all the segmentation modes, selecting the segmentation mode with the largest comprehensive score as an optimal segmentation result, and taking each segment in the segmentation mode as an optimal segment of an independent filling table.
7. The AI-converged intelligent manufacturing plant instruction processing output method of claim 6, wherein the calculation of the independent score includes: For any word in the segment, calculating cosine similarity between the word vector of the word and the word vector of each attribute in the best-matched second event table to obtain a similarity sequence, and calculating standard deviation of the similarity sequence as an independent score of the word.
8. The AI-fused intelligent manufacturing plant instruction processing output method of claim 1, wherein populating the obtained optimal segment into the second event table and outputting comprises: For any optimal segment, calculating cosine similarity between a word vector of any word in the optimal segment and word vectors of all the special terms in the special word stock, and selecting the special term with the maximum cosine similarity as a standardized mapping result of the word; Filling each mapped technical term into a corresponding attribute position according to the word vector similarity of each attribute in the second event table to form a complete event record; and outputting the filled second event table to the manufacturing system as a result of the second event table of the current identification result.
9. The AI-converged intelligent manufacturing plant instruction processing output method of claim 1, wherein the table types include at least a patrol record table and a fault registry table, wherein the patrol record table includes three attributes of a device name, a patrol state, and a patrol time, and the fault registry table includes three attributes of a device name, a fault phenomenon, and an occurrence time.
An AI-fused intelligent manufacturing shop instruction processing output system comprising a processor and a memory, the memory storing computer program instructions that when executed by the processor implement the AI-fused intelligent manufacturing shop instruction processing output method of any one of claims 1-9.

Description

AI-fused intelligent manufacturing workshop instruction processing output method and system Technical Field The present invention relates to the field of data processing. More particularly, the invention relates to an AI-fused intelligent manufacturing shop instruction processing output method and system. Background In an intelligent manufacturing workshop, processing of data such as production instructions, inspection records, fault registration and the like mainly depends on manual input of workers, such as filling in paper documents or operating terminal equipment such as a tablet, a keyboard and the like. However, workshop workers generally have weaker software operation capability, manual input is long in time consumption and easy to make mistakes, so that problems of data delay and secondary input are caused, information circulation among systems such as production, quality and equipment is not smooth, fault maintenance or production adjustment depends on manual judgment, reaction time is long, and overall efficiency and safety are affected. To reduce the burden of manual entry, the prior art attempts to introduce automatic speech recognition (ASR, automatic Speech Recognition) techniques to match form types by capturing the worker's speech and converting it to text, and then directly calculating the word vector semantic similarity between the white words and the form attributes. However, the method ignores a large number of non-form attribute words commonly existing in the manufacturing instruction, such as turning words, personal words, language words and the like, and the words are irrelevant to form attributes but participate in similarity calculation, so that the form types are easily dominated by the non-attribute words when matched, and the type matching is wrong, and then instruction information is transmitted to wrong event processing flow, so that information transmission abnormality is caused. Disclosure of Invention In order to solve the technical problem that the form type matching is wrong and the event processing flow is abnormal due to the fact that a large number of non-attribute words exist in the spoken language instruction in the prior art, the invention provides the following aspects. In a first aspect, an AI-converged intelligent manufacturing plant instruction processing output method includes: Acquiring a history sample library which is accumulated in a history manufacturing stage and comprises a form type, a professional word library, a first event form recorded as an original spoken language instruction form, a recognition result and a second event form recorded as a professional term, and acquiring a current recognition result obtained by ASR (automatic repeat request) conversion of the worker voice in the current manufacturing stage; Calculating the definition of each attribute in each form type based on the historical sample library and the current recognition result, matching the definition to obtain a second event form corresponding to the current recognition result, and dividing the current recognition result based on the matched second event form to obtain an optimal segment of the independent form filling; filling the obtained optimal fragment into a second event table and outputting the optimal fragment; The definition is comprehensively calculated based on the difference between the white words in the first event table, the difference between the proprietary terms in the second event table and the semantic proximity between the white words and the proprietary terms under each attribute; The step of matching the second event table corresponding to the current identification result by using the definition comprises the following steps: Carrying out semantic clustering on the current recognition result, respectively calculating the association strength between the cluster and the attribute from the history data association dimension and the attribute definition association dimension between each cluster and each attribute, combining the definition of each attribute, comprehensively calculating the event relativity between the current recognition result and each table type, and selecting the table type with the largest event relativity as a best-matched second event table. Optionally, the calculating of the sharpness includes: For any attribute in any form type, acquiring all white words corresponding to the attribute in a first event form in a history sample library, acquiring all special terms corresponding to the attribute in a second event form in the history sample library, and respectively calculating a first characteristic value, a second characteristic value and a standard degree of the corresponding attribute based on the white words and the special terms; And calculating the difference value between the second characteristic value and the first characteristic value, converting the difference value between the second characteristic value and the first character