Intelligent manufacturing workshop instruction processing output method and system based on AI fusion

Abstract

This invention relates to the field of data processing, and more specifically, to an AI-integrated intelligent manufacturing workshop instruction processing and output method and system. The method includes: acquiring a historical sample library accumulated from previous manufacturing stages, and collecting the current recognition result obtained by ASR conversion of worker speech at the current manufacturing stage; calculating the clarity of each attribute in each table type based on the historical sample library and the current recognition result; matching the clarity to a second event table corresponding to the current recognition result; segmenting the current recognition result based on the matched second event table to obtain the optimal segment for independent form filling; filling the obtained optimal segment into the second event table and outputting it. This invention effectively suppresses the interference of non-attribute words in spoken instructions on table type matching, avoiding information transmission anomalies caused by incorrect matching leading to the transmission of instruction information to the error handling process.

Description

Technical Field

[0001] This invention relates to the field of data processing. More specifically, this invention relates to a method and system for processing and outputting instructions in an AI-integrated intelligent manufacturing workshop. Background Technology

[0002] In smart manufacturing workshops, the processing of data such as production instructions, inspection records, and fault registrations mainly relies on manual input by workers, such as filling out paper forms or operating terminal devices like tablets and keyboards. However, workshop workers generally have weak software operation skills, and manual input is time-consuming and prone to errors, leading to data delays and secondary input problems. This results in poor information flow between production, quality, and equipment systems, and fault repair or production adjustments rely on human judgment, leading to long response times and impacting overall efficiency and safety.

[0003] To reduce the burden of manual data entry, existing technologies attempt to introduce Automatic Speech Recognition (ASR). This involves collecting workers' speech, converting it into text, and then directly calculating the semantic similarity between colloquial words and table attributes to match table types. However, this method ignores a large number of non-table attribute words commonly found in manufacturing instructions, such as transition words, personal pronouns, and modal particles. These words are unrelated to table attributes but participate in similarity calculations, causing table type matching to be dominated by non-attribute words, resulting in type matching errors. Consequently, instruction information is transmitted to incorrect event processing flows, leading to information transmission anomalies. Summary of the Invention

[0004] To address the technical problems in the prior art where the presence of numerous non-attribute words in spoken instructions leads to table type matching errors and abnormal event processing flows, this invention provides solutions in the following aspects.

[0005] In the first aspect, the AI-integrated intelligent manufacturing workshop instruction processing and output method includes:

[0006] Acquire a historical sample library accumulated from the historical manufacturing stages, which includes table types, professional lexicons, first event tables recorded as original spoken instructions, recognition results, and second event tables recorded as professional terms; and collect the current recognition results obtained by converting the speech of workers in the current manufacturing stage into ASR.

[0007] Based on the historical sample database and the current recognition results, the clarity of each attribute in each table type is calculated. The clarity is used to match the second event table corresponding to the current recognition result. Based on the matched second event table, the current recognition result is segmented to obtain the optimal segment for independent form filling.

[0008] Fill the second event table with the obtained optimal fragment and output it;

[0009] The clarity is calculated based on the differences between colloquial words in the first event table under each attribute, the differences between technical terms in the second event table, and the semantic similarity between colloquial words and technical terms.

[0010] The table of second events corresponding to the current recognition result, matched using sharpness, includes:

[0011] Semantic clustering is performed on the current identification results. The association strength between clusters and attributes is calculated from the historical data association dimension and the attribute definition association dimension of each cluster. Combined with the clarity of each attribute, the event relevance between the current identification results and each table type is calculated. The table type with the highest event relevance is selected as the best matching second event table.

[0012] Optionally, the calculation of the sharpness includes:

[0013] For any attribute in any table type, obtain all plain language words corresponding to that attribute in the first event table of the historical sample library, obtain all professional terms corresponding to that attribute in the second event table of the historical sample library, and calculate the first feature value, second feature value and standard degree of the corresponding attribute based on the plain language words and professional terms respectively.

[0014] Calculate the difference between the second eigenvalue and the first eigenvalue. Convert the difference between the second eigenvalue and the first eigenvalue into an exponential value using an exponential function, and then multiply it by the standard value. The resulting product is the sharpness of the attribute.

[0015] Optionally, the calculation of the first eigenvalue includes:

[0016] Obtain the word vectors of each colloquial word, calculate the cosine similarity between any two word vectors, then calculate the negative exponent value of each cosine similarity, and take the mean of all negative exponent values ​​as the first feature value.

[0017] Optionally, the calculation of the second eigenvalue includes:

[0018] Obtain the word vectors of each technical term, and calculate the second feature value using the same method as the first feature value calculation.

[0019] Optionally, the calculation of the standard degree includes:

[0020] The mean cosine similarity between the word vectors of each colloquial word and the corresponding technical term is calculated and used as the standard.

[0021] Optionally, the current recognition result is segmented based on the matched second event table to obtain the optimal segment for independent form filling, including:

[0022] The current recognition result is segmented to obtain a word sequence. The word sequence is then randomly divided to generate multiple segmentation methods. Each segmentation method divides the word sequence into several continuous segments, and the shortest length of each segment is not less than the number of attributes in the best-matching second event table.

[0023] For any segment under any segmentation method, calculate the independent score of each word in the segment, and calculate the mean of the independent scores of all words in the segment;

[0024] The sum of the Euclidean distances between all pairs of similarity sequences of words within the segment is used as the segment's partition score.

[0025] Add the mean of the independent scores within the segment to the segmentation score, and then take the average of all segments under this segmentation method to obtain the comprehensive score of this segmentation method;

[0026] Iterate through all segmentation methods, select the segmentation method with the highest comprehensive score as the optimal segmentation result, and take each segment under this segmentation method as the optimal segment for independent table filling.

[0027] Optionally, the calculation of independent fractions includes:

[0028] For any word in the segment, calculate the cosine similarity between the word vector of the word and the word vectors of each attribute in the best matching second event table to obtain a similarity sequence. Calculate the standard deviation of the similarity sequence as the independent score of the word.

[0029] Optionally, the obtained optimal fragment is populated into the second event table and the output includes:

[0030] For any optimal segment, calculate the cosine similarity between the word vector of any word in the optimal segment and the word vectors of each professional term in the professional thesaurus, and select the professional term with the largest cosine similarity as the standardized mapping result of the word.

[0031] The mapped technical terms are filled into the corresponding attribute positions according to their word vector similarity with each attribute in the second event table to form a complete event record;

[0032] The completed second event table is output to the manufacturing system as the result of the second event table of the current identification results.

[0033] Optionally, the table types include at least an inspection record table and a fault registration table. The inspection record table contains three attributes: equipment name, inspection status, and inspection time. The fault registration table contains three attributes: equipment name, fault phenomenon, and occurrence time.

[0034] Secondly, an AI-integrated intelligent manufacturing workshop instruction processing and output system includes: a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the AI-integrated intelligent manufacturing workshop instruction processing and output method described in any one of the above claims is implemented.

[0035] The present invention has the following beneficial effects:

[0036] 1. This invention calculates the clarity of each attribute, which integrates the differences between colloquial words in the first event table, the differences between technical terms in the second event table, and the semantic similarity between colloquial words and technical terms, using this as the quantitative basis for attribute weights. Based on this, semantic clustering is performed on the current recognition results. The association strength between clusters and attributes is calculated from both the historical data association dimension and the attribute definition association dimension. The event relevance is then comprehensively calculated by combining the clarity of each attribute, and the table type with the highest event relevance is selected as the best-matching second event table. This dual association mechanism effectively suppresses the interference of non-attribute words (such as modal particles, personal pronouns, and transition words) in spoken instructions on table type matching, avoiding information transmission anomalies caused by incorrect matching leading to the transmission of instruction information to the error handling process.

[0037] 2. To address the common problem of mixed information from multiple devices and events in spoken instructions in workshops, this invention segments the current recognition result based on a matched second event table. It calculates the independent score of words and the segmentation score of segments. The independent score reflects the standard deviation of the similarity sequence between a word and the word vectors of each attribute, while the segmentation score reflects the cumulative Euclidean distance of the similarity sequences between all pairs of words within a segment. A comprehensive score for the segmentation method is obtained by combining both scores, and the segmentation method with the highest comprehensive score is selected as the optimal segment for independent table filling. This segmentation mechanism takes into account both word-level semantic directionality and segment-level semantic differences, effectively solving the problems of information confusion and multiple event overlap during whole-sentence processing. Combined with a professional lexicon, the colloquial words in the optimal segment are mapped to professional terms with the highest cosine similarity, and filled into the corresponding attribute positions according to the similarity with the word vectors of each attribute, forming a structured event record. This achieves end-to-end automated conversion from unstructured speech to standardized tabular data. Attached Figure Description

[0038] Figure 1 This is a flowchart of steps S1-S3 in the AI-integrated intelligent manufacturing workshop instruction processing and output method of this invention.

[0039] Figure 2 This is a schematic diagram illustrating the operation of segmenting the current recognition result into independent form-filling segments in the AI-integrated intelligent manufacturing workshop instruction processing and output method of this invention. Detailed Implementation

[0040] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0041] This invention is applied to a smart manufacturing workshop scenario. Workers in the workshop have varying skill levels and often use colloquial language for vague or non-standard expressions. The workshop is equipped with a microphone array to collect the workers' voices.

[0042] Reference Figure 1 The AI-integrated intelligent manufacturing workshop instruction processing and output method includes steps S1-S3, as follows:

[0043] S1: Acquire a historical sample library accumulated from the historical manufacturing stages, including table types, professional lexicons, and a first event table, recognition results, and a second event table, and collect the current recognition results obtained by converting the worker's speech in the current manufacturing stage through ASR.

[0044] To enable the system to adapt to the language habits of the workers in this workshop and achieve personalized and accurate matching, the system first acquires data accumulated from previous manufacturing stages to build a historical sample database. This historical data records the workers' past spoken instructions and their corresponding standard forms.

[0045] The system predefined includes at least:

[0046] 1. Table Types: At least "Inspection Record Form" and "Fault Registration Form" are required. The "Inspection Record Form" includes three attributes: "Equipment Name," "Inspection Status," and "Inspection Time." The "Fault Registration Form" includes three attributes: "Equipment Name," "Fault Phenomenon," and "Time of Occurrence."

[0047] 2. Specialized terminology database: Stores the mapping relationship between colloquial words and professional terms. For example, "normal" is mapped to "running normally", "can" is mapped to "running normally", "no problem" is mapped to "running normally", "needs to be repaired" is mapped to "awaiting repair", "broken" is mapped to "fault", etc.

[0048] 3. Historical Manufacturing Stage Data: This includes 1000 historical recognition results accumulated over the past three months, along with their corresponding first and second event tables. The first event table refers to the original spoken instructions, reflecting the workers' most natural descriptive style. The recognition results are the text statements obtained after the workers' spoken instructions have been transcribed by an Automatic Speech Recognition (ASR) system. During the transcription process, the ASR system converts spoken instructions into expressions closer to written language; therefore, the recognition results are usually in professional or semi-professional language, rather than raw colloquial speech. The second event table refers to the table records filled in after the colloquial expressions in the first event table have been converted into standardized professional terminology. This is a standard format that the system can recognize and is ultimately output to the manufacturing system.

[0049] The stored data—table types, specialized terminology, and historical manufacturing stage data—together constitute the historical sample library.

[0050] In addition, it is necessary to capture the worker's spoken speech through a microphone, and then process it with ASR to obtain the current recognition result.

[0051] This completes the loading of the historical sample database and the acquisition of the current recognition results, providing a data foundation for subsequent core calculations.

[0052] S2: Based on the historical sample library and the current recognition result, calculate the clarity of each attribute in each table type, use the clarity to match the second event table corresponding to the current recognition result, and segment the current recognition result based on the matched second event table to obtain the optimal segment for independent form filling.

[0053] After data preparation, the spoken instructions uttered by workers during the current manufacturing phase need to be converted into structured tabular data. However, there are two main challenges with workers' spoken expressions: first, the degree of ambiguity varies significantly across different attributes (e.g., equipment names are clear, but equipment status is ambiguous), making direct matching prone to errors; second, spoken instructions often contain non-attribute words and may mix information from multiple devices, leading to information chaos when processed as a whole sentence. Therefore, this embodiment of the invention first quantifies the degree of ambiguity of each attribute in each table type in the historical sample database as a weight, then accurately matches the most suitable second event table (i.e., the table filled with standardized professional terms) based on the weight, and finally segments the current recognition result into independently filled fragments, thus preparing for subsequent terminology conversion and output.

[0054] Reference Figure 2 The operation of segmenting the current recognition result into independent form-filling segments includes steps S20-S22:

[0055] S20: Calculate the clarity of each attribute in each table type in the historical sample database.

[0056] Taking an inspection record table as an example, with the inspection status as the attribute in the table type, extract plain language terms (normal, okay, no problem, running well, etc.) about the inspection status from all first event tables in the historical manufacturing stage, as well as professional terms (such as running normally, running well, etc.) about the inspection status from the corresponding second event tables.

[0057] Furthermore, the Word2Vec model is used to obtain standardized word vectors of the colloquial terms representing the inspection status in the first event table. The cosine similarity between any two word vectors is calculated, and then the negative exponent of the cosine similarity between any two word vectors (e.g., the negative cosine similarity power of the natural constant e) is calculated. The mean of all negative exponent values ​​of cosine similarity is used as the first feature value of the inspection status. This first feature value reflects the degree of difference between the colloquial terms used by workers to describe the inspection status attribute in the inspection record table. A larger negative exponent value indicates a greater difference in word vectors, less similar the colloquial terms, and higher and clearer the distinguishability of the attribute at the colloquial level.

[0058] The operation of calculating the first feature value is performed on the second event table to obtain the corresponding second feature value of the inspection status. The second feature value reflects the degree of difference between the professional terms of the inspection status attributes of the inspection record table. The smaller the negative exponent value, the more similar the professional terms are, and the more unified the professional level is.

[0059] Furthermore, the cosine similarity of word vectors between the colloquial terms of the inspection status in each first event table and the corresponding technical terms in each second event table is calculated under the inspection status attribute. The mean of all cosine similarities calculated here is used as the standard score of the inspection status attribute. The higher the standard score, the closer the colloquial terms are to the semantics of the technical terms, the more standard the worker's expression, and the clearer the colloquial expression of this attribute.

[0060] Furthermore, based on the first feature value, second feature value, and standard degree obtained above, the clarity of the inspection status attribute in the inspection record table is calculated. Specifically, the difference between the second feature value and the first feature value is calculated, and then the difference between the second feature value and the first feature value is converted into an exponential value through an exponential function and multiplied by the product of the standard degree of the inspection status. The result of the multiplication is taken as the clarity of the inspection status.

[0061] The method measures the relative magnitude of differences between specialized terms and between colloquial terms by calculating the difference between the second and first eigenvalues. A larger difference indicates a greater distinction between colloquial terms relative to the distinction between specialized terms. This means workers can use distinctly different colloquial terms to differentiate between concepts with minor professional differences. In this case, colloquial terms have stronger distinguishing and recording capabilities, and clarity should be significantly increased. Furthermore, the difference between the second and first eigenvalues ​​is transformed using an exponential function. This is because the exponential function has a monotonically increasing characteristic; a positive difference amplifies the effect, while a negative difference compresses it, and the range is positive, avoiding zero or negative values.

[0062] Based on the above operations, the clarity of each attribute in the inspection record table can be obtained. Similarly, the clarity of each attribute in other table types, such as the fault registration table, can also be obtained.

[0063] S21: Use the clarity of the attributes to match the second event table corresponding to the current recognition result.

[0064] The spoken language produced by workers during the current manufacturing phase, after being converted by ASR (Automatic Speech Recognition), often contains a large number of non-attributive words (such as modal particles, personal pronouns, and transition words), and may also describe multiple devices or events. Directly calculating the global semantic similarity between the entire sentence and the table is easily interfered with by irrelevant words, leading to incorrect matching of the second event table. Therefore, it is necessary to perform semantic clustering on the words in the current recognition results, separate the core descriptive words, and combine them with the attribute clarity weights obtained from S20 to accurately determine the second event table that best matches the current recognition results.

[0065] First, the jieba word segmentation tool is used to segment the current recognition result to obtain a word sequence, for example: [Equipment Inspection Report, Today, the milling machine is running normally, the lathe needs maintenance].

[0066] Next, Word2Vec is used to obtain the standardized word vector of each word, and K-means clustering is performed using the cosine similarity between word vectors as the distance metric to obtain multiple clusters.

[0067] Then, taking the inspection record table as an example again, with the inspection status as the attribute in the table, and selecting any cluster as the target cluster, we calculate the cosine similarity between the word vectors of each word in the target cluster and the word vectors of all plain language words related to the inspection status in the first event table, and calculate the average to obtain the first similarity between the target cluster and the inspection status. This first similarity measure quantifies the overall association strength between a cluster and all historical data entered under any attribute in the inspection record table.

[0068] Furthermore, the product of the maximum first similarity between any attribute and all clusters in the inspection record table and the clarity of that attribute is calculated. This product is then summed across all attributes, and the sum is used as the correlation factor between the current identification result and the inspection record table. This correlation factor reflects the overall semantic association between the colloquial words in the current identification result and the content of historical records, avoiding the dominance of a single broad attribute semantics in the matching result.

[0069] Next, the cosine similarity between the word vectors of each word in the target cluster and the word vectors of the inspection state is calculated. Then, the calculated cosine similarities for all word pairs in the target cluster are summed and averaged to obtain the second similarity between the target cluster and the inspection state. This second similarity, in turn, yields the second similarity between the target cluster and any attribute in the inspection record table. This second similarity measures the direct association strength between a cluster and a single attribute, without relying on historical data, reflecting the degree of matching only at the attribute definition level.

[0070] Then, the standard deviation of the second similarity between the target cluster and all attributes in the inspection record table is calculated to obtain the correlation between the target cluster and the inspection record table. This standard deviation reflects the fluctuation of the association strength between the target cluster and each attribute in the inspection record table. The larger the standard deviation, the stronger the semantic association between the target cluster and some attributes in the inspection record table, while the association with other attributes is very weak. This reflects the specific matching between the cluster and the attribute structure. Conversely, the smaller the standard deviation, the more even the semantic association between the target cluster and all attributes, and the lower the distinguishability.

[0071] Finally, the mean correlation between all clusters and the inspection record table is multiplied by the correlation factor between the current identification result and the inspection record table, and the result of the product is taken as the event correlation between the current identification result and the inspection record table.

[0072] For each table type, the event relevance to the current recognition result is calculated according to the above steps. Finally, the historical data of the table type with the highest event relevance is selected as the second event table that best matches the current recognition result.

[0073] It should be noted that during the matching phase, the selected "second event table" actually refers to a table type (such as the structure of an inspection record table), and at this time, the second event table has not yet been filled with specific content.

[0074] S22: Segment the current recognition result based on the matched second event table to obtain the optimal segment for independent form filling.

[0075] In real-world manufacturing scenarios, the spoken instructions output by workers, after being converted by ASR (Automatic Speech Recognition), often form long text sequences with random word order, which may contain semantic information describing multiple devices or events. Directly matching the entire instruction can easily lead to information confusion. Therefore, this invention, based on the aforementioned second event table that yields the best match, further performs optimal segmentation on the current recognition results to extract the optimal fragment corresponding to a single device or event, providing a data foundation for subsequent accurate recording. The specific process is as follows:

[0076] First, the word sequence of the current recognition result is obtained, and the word sequence is randomly segmented to generate multiple segmentation methods. Each segmentation method divides the original word sequence into several continuous segments. To ensure that each segment has sufficient information to complete the table filling, the minimum length of each segment is set to be no less than the number of attributes of the second event table that best matches the current recognition result.

[0077] Secondly, for any segmentation method, select one segment as the target segment. For any word in the target segment as the target word, calculate the cosine similarity (or Pearson correlation coefficient) between the word vector of the target word and the word vector of each attribute in the best-matching second event table, thus obtaining the similarity sequence of the target word. The length of this similarity sequence is equal to the number of attributes in the best-matching second event table.

[0078] Furthermore, the standard deviation of the similarity sequence of the target words is calculated as the independent score of the target words. Calculating the independent score can reflect the difference in the degree of association between the target word and each attribute of the most matching second event table: the larger the standard deviation, the stronger the association between the target word and some attributes and the weaker the association with other attributes, that is, the target word has a clear semantic direction; conversely, the smaller the standard deviation, the more even the association between the target word and each attribute, and the semantics are relatively ambiguous.

[0079] Next, for any segment under a given segmentation method, the mean of the independent scores of all words within that segment is calculated. Simultaneously, the Euclidean distance between each pair of similarity sequences of all words within that segment is calculated, and the Euclidean distances of all word pairs are summed to obtain the segment's segmentation score. The segmentation score reflects the semantic differences between words within a segment: a higher segmentation score indicates greater differences in the attribute structures pointed to by the words within the segment, meaning that the segment may contain mixed information describing multiple different devices or events.

[0080] Furthermore, for a segmentation method, the average of the segmentation scores and independent scores of all its constituent segments is added together, and then the average of all segments is taken to obtain the comprehensive score of that segmentation method. The addition of the average segmentation scores and independent scores achieves a complementary effect: a word can independently point to an attribute (i.e., a high independent score), while the differences between words within the same segment can also be significant (i.e., a high segmentation score). Addition ensures that the contributions of both dimensions are cumulative. A segment needs to simultaneously satisfy both conditions—"each word has a clear affiliation" and "significant differences between words"—to obtain a high score.

[0081] Finally, all the segmentation methods generated above are traversed, and their comprehensive scores are calculated respectively. The segmentation method with the highest comprehensive score is selected as the optimal segmentation result of the current recognition result, and each segment under this segmentation method (i.e. the optimal segment) is used as the input for subsequent event recording processing. Each segment corresponds to an independent event recording unit.

[0082] S3: Fill the second event table with the obtained optimal fragment and output it.

[0083] In S2 above, the most suitable event table type (i.e., the second event table) has been matched for the current recognition result, and the current recognition result has been optimally segmented to obtain the optimal segment corresponding to a single device or a single event.

[0084] Furthermore, it is necessary to accurately map the colloquial words in these optimal fragments to the corresponding attribute positions in the best-matching second event table, thereby automatically populating the event records and outputting the results to the manufacturing system for subsequent use. The specific process is as follows:

[0085] The optimal segmentation result of the current recognition is obtained, i.e., one or more optimal segments. For any optimal segment, each colloquial word in the segment is extracted as the target word. The cosine similarity between the word vector of the target word and the word vector of each professional term in the professional terminology database is calculated, resulting in a series of similarity values. The professional term with the highest cosine similarity is selected as the standardized mapping result corresponding to the target word. This matching process ensures that even if workers use colloquial or non-standard expressions, they can be accurately mapped to professional terms that the system can recognize.

[0086] Next, for the event corresponding to the current optimal segment, the mapped technical terms are filled into the corresponding attribute positions of the second event table according to their semantic relevance. Specifically, by calculating the word vector similarity between each mapped technical term and each attribute of the second event table, the technical term is stored in the attribute column with the highest similarity. The above operation is repeated until all colloquial words in the current optimal segment have been mapped and filled, forming a complete event record.

[0087] Finally, the completed second event table is output to the manufacturing system as the result of the current identification results. The manufacturing system performs corresponding production scheduling, equipment control, or anomaly handling operations based on the content recorded in the second event table, achieving rapid response and accurate recording of shop floor events.

[0088] The present invention also provides an AI-integrated intelligent manufacturing workshop instruction processing and output system. The system includes a processor and a memory, the memory storing computer program instructions. When the computer program instructions are executed by the processor, the AI-integrated intelligent manufacturing workshop instruction processing and output method according to the first aspect of the present invention is implemented.

[0089] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

[0090] It should be noted that those skilled in the art can make various modifications and improvements without departing from the inventive concept, and these all fall within the scope of protection of this invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. An AI-fused intelligent manufacturing plant instruction processing output method, characterized by, include: Acquire a historical sample library accumulated from the historical manufacturing stages, which includes table types, professional lexicons, first event tables recorded as original spoken instructions, recognition results, and second event tables recorded as professional terms; and collect the current recognition results obtained by converting the speech of workers in the current manufacturing stage into ASR. Based on the historical sample database and the current recognition results, the clarity of each attribute in each table type is calculated. The clarity is used to match the second event table corresponding to the current recognition result. Based on the matched second event table, the current recognition result is segmented to obtain the optimal segment for independent form filling. Based on the matching second event table, the current recognition result is segmented to obtain the optimal segment for independent form filling, including: The current recognition result is segmented to obtain a word sequence. The word sequence is then randomly divided to generate multiple segmentation methods. Each segmentation method divides the word sequence into several continuous segments, and the shortest length of each segment is not less than the number of attributes in the best-matching second event table. For any segment under any segmentation method, calculate the independent score of each word in the segment, and calculate the mean of the independent scores of all words in the segment; The sum of the Euclidean distances between all pairs of similarity sequences of words within the segment is used as the segment's partition score. Add the mean of the independent scores within the segment to the segmentation score, and then take the average of all segments under this segmentation method to obtain the comprehensive score of this segmentation method; Iterate through all segmentation methods, select the segmentation method with the highest comprehensive score as the optimal segmentation result, and take each segment under this segmentation method as the optimal segment for independent table filling; The calculation of independent fractions includes: For any word in the segment, calculate the cosine similarity between the word vector of the word and the word vectors of each attribute in the best matching second event table to obtain a similarity sequence, and calculate the standard deviation of the similarity sequence as the independent score of the word. Fill the second event table with the obtained optimal fragment and output it; The clarity is calculated based on the differences between colloquial words in the first event table under each attribute, the differences between technical terms in the second event table, and the semantic similarity between colloquial words and technical terms. The table of second events corresponding to the current recognition result, matched using sharpness, includes: Semantic clustering is performed on the current identification results. The association strength between clusters and attributes is calculated from the historical data association dimension and the attribute definition association dimension of each cluster. Combined with the clarity of each attribute, the event relevance between the current identification results and each table type is calculated. The table type with the highest event relevance is selected as the best matching second event table. 2.The AI-fusion smart manufacturing plant instruction processing output method of claim 1, characterized in that, The calculation of the sharpness includes: For any attribute in any table type, obtain all plain language words corresponding to that attribute in the first event table of the historical sample library, obtain all professional terms corresponding to that attribute in the second event table of the historical sample library, and calculate the first feature value, second feature value and standard degree of the corresponding attribute based on the plain language words and professional terms respectively. Calculate the difference between the second eigenvalue and the first eigenvalue. Convert the difference between the second eigenvalue and the first eigenvalue into an exponential value using an exponential function, and then multiply it by the standard value. The resulting product is the sharpness of the attribute.

3. The AI-integrated intelligent manufacturing workshop instruction processing and output method according to claim 2, characterized in that, The calculation of the first eigenvalue includes: Obtain the word vectors of each colloquial word, calculate the cosine similarity between any two word vectors, then calculate the negative exponent value of each cosine similarity, and take the mean of all negative exponent values ​​as the first feature value.

4. The AI-integrated intelligent manufacturing workshop instruction processing and output method according to claim 3, characterized in that, The calculation of the second eigenvalue includes: Obtain the word vectors of each technical term, and calculate the second feature value using the same method as the first feature value calculation.

5. The AI-integrated intelligent manufacturing workshop instruction processing and output method according to claim 4, characterized in that, The calculation of standard degree includes: The mean cosine similarity between the word vectors of each colloquial word and the corresponding technical term is calculated and used as the standard.

6. The AI-integrated intelligent manufacturing workshop instruction processing and output method according to claim 1, characterized in that, The obtained optimal fragment is populated into the second event table and the output includes: For any optimal segment, calculate the cosine similarity between the word vector of any word in the optimal segment and the word vectors of each professional term in the professional thesaurus, and select the professional term with the largest cosine similarity as the standardized mapping result of the word. The mapped technical terms are filled into the corresponding attribute positions according to their word vector similarity with each attribute in the second event table to form a complete event record; The completed second event table is output to the manufacturing system as the result of the second event table of the current identification results.

7. The AI-integrated intelligent manufacturing workshop instruction processing and output method according to claim 1, characterized in that, The table types include at least an inspection record form and a fault registration form. The inspection record form contains three attributes: equipment name, inspection status, and inspection time. The fault registration form includes... It includes three attributes: equipment name, fault symptoms, and time of occurrence.

8. An AI-integrated intelligent manufacturing workshop instruction processing and output system, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, which, when executed by the processor, implement the AI-integrated intelligent manufacturing workshop instruction processing and output method according to any one of claims 1-7.

Images