Aero-engine blade identification code reading result post-processing review correction method

Abstract

This invention relates to a post-processing review and correction method for the identification results of aero-engine blades, belonging to the field of industrial character recognition technology. First, the blade identification code identification result is input. Based on the Trie data structure and referencing an existing sample library containing all blade identification codes, the accuracy of the identification result is judged by a post-processing review method. If correct, no post-processing correction is performed; if incorrect, the input blade identification code identification result is corrected according to the post-processing correction method provided by this invention, and the final correct identification result is output through interactive selection. Through the application of this invention, the review and correction of aero-engine blade identification code identification results can be realized, achieving a high overall identification accuracy rate, avoiding a large-scale manual search and verification process, and providing a foundation for the informatization and intelligent upgrading of the blade's entire life cycle management process.

Description

Technical Field

[0001] This invention belongs to the field of industrial character recognition technology, and in particular relates to a method for post-processing review and correction of industrial identification code recognition results. Background Technology

[0002] To achieve comprehensive management of aero-engine blades throughout their entire lifecycle, blades need to be directly identified for identification and traceability. This is typically accomplished by encoding and reading the blade tenons. Currently, manual reading is largely used to address the needs of industrial identification coding. The low contrast of the parts' metallic color and complex environmental interference increase the difficulty of reading, and large-scale production of parts incurs significant labor costs. These problems are even more pronounced with blades requiring selective assembly, increasing worker visual fatigue and mental strain, and easily leading to errors. Researching semi-automatic coding and reading technology, supplemented by minimal human-machine interaction correction, to replace traditional manual reading can help improve production efficiency and ensure quality.

[0003] For critical aerospace products like aero engines, which require high safety and reliability, any error can have serious consequences. Similar characters and low-quality characters can both lead to misidentification, and the structured reading process for discrete character recognition results may introduce problems such as missed readings, out-of-order readings, and serial readings. Therefore, it is difficult to avoid misidentification during the reading of aero engine blade identification codes. Researching post-processing review and correction techniques for structured reading results of blade identification codes can correct errors that occur during the reading process, improve the accuracy of the reading results, and thus ensure the security of the reading data. At the same time, semi-automatic detection and correction of reading errors can reduce the burden of manual review, improve work efficiency, and ensure the quality of reading results. However, existing post-processing review methods are mostly based on direct matching and searching in databases, which is extremely time-consuming when dealing with large amounts of data; existing post-processing correction methods do not fully utilize data features and are mostly based on a single sample library, which cannot effectively correct misidentifications. Summary of the Invention

[0004] The technical problem to be solved by this invention is:

[0005] To overcome the shortcomings of existing technologies, this invention provides a method for post-processing review and correction of aero-engine blade identification code reading results. By introducing a Trie data structure, efficient post-processing review of the reading results is achieved. A post-processing correction candidate result push method based on multi-parameter evaluation and probability calculation is employed, and human-machine collaboration is used for post-processing correction to ensure the accuracy and reliability of the final reading results.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0007] A method for post-processing review and correction of aircraft engine blade identification code reading results, characterized by the following steps:

[0008] Step 1: Utilize the characteristics of the Trie structure to construct a complete Trie structure for the coding sample library, and perform post-processing review on the leaf identification coding recognition results to determine the accuracy of the recognition results;

[0009] Step 2: Post-processing correction is performed on the coding error identification results. Based on the idea of ​​dynamic programming, the similarity between the error identification code and the sample library code is calculated to obtain the post-processing correction candidate set. Dynamic programming is used to solve the edit distance between the error identification code and the post-processing candidate set, and the Bayes probability when the edit distance is 1 is calculated. Based on the probability ranking method of the post-processing candidate set based on similarity, edit distance and Bayes probability values, the code with the highest probability is selected to complete the post-processing correction of the error identification results.

[0010] A further technical solution of the present invention: Step 1 includes:

[0011] Step 1.1: Construct the Trie structure of the complete encoded sample library. Read the codes in the encoded sample library one by one. Each code is a key of the Trie structure, and the characters of the code are child nodes of the key. Construct the Trie structure of the complete encoded sample library according to the requirements of the Trie structure.

[0012] Step 1.2: Accuracy judgment of blade code recognition results based on Trie structure: Read the code characters to be reviewed in sequence and determine whether the characters exist in the child nodes of the Trie structure of the complete code sample library. If they do not exist, it is an incorrect recognition. If they exist, move to the next Trie child node to judge the next code character until the last character of the code to be reviewed is judged. If all characters of the code to be reviewed exist in the Trie child nodes, the recognition is correct.

[0013] A further technical solution of the present invention: Step 2 includes:

[0014] Step 2.1: Calculate the similarity between the error code recognition result and the sample library code, and obtain a list of similarity values;

[0015] Step 2.2: Sort the similarity list according to the size of the similarity value, and select the top n sample database codes to form a post-processing correction candidate set according to the actual task requirements;

[0016] Step 2.3: Calculate the edit distance between the error identification result and the post-processing correction candidate set to obtain a list of edit distance values;

[0017] Step 2.4: Calculate the Bayesian probability when the edit distance is 1, and then calculate the P(c) probability for all candidates with an edit distance of 1. c |c i By sorting the values ​​from largest to smallest, the best candidate correction identifier code can be obtained;

[0018] Step 2.5: Comprehensive ranking of candidate results: Based on the data obtained in the above steps, the post-processing correction candidate set is first divided into two categories according to the edit distance: edit distance 1 and the other two categories. The category with edit distance 1 is ranked according to Bayesian probability, and the other categories are directly ranked according to similarity. The category with edit distance 1 has higher priority than the other categories. The probability ranking of the candidate results is obtained by comprehensive ranking, and the candidate code with the highest probability at the top is output as the correction result.

[0019] A further technical solution of the present invention: the calculation of the similarity between the error code recognition result and the sample library code specifically includes:

[0020]

[0021] Wherein, C(X) m ,Y n ) represents X m and Y n The number of elements in the longest common subsequence of sequence X, where m and n represent the sequence X. m and Y n The number of elements, where m and n are 0, represents X. m and Y n An empty sequence and Representing sequence X m and Y n The number of elements.

[0022] A further technical solution of the present invention: the calculation of the edit distance between the error identification result and the post-processing correction candidate set specifically includes:

[0023]

[0024] Wherein, lev(X) m ,Y n ) represents X m and Y n Edit distance.

[0025] A further technical solution of the present invention: the calculation of the Bayesian probability when the edit distance is 1 is specifically as follows:

[0026]

[0027]

[0028]

[0029] Wherein, P(c c |c i ) indicates that the input character is c i After correction, the character is 'c'. c The probability, P(c) i |c c ) indicates the correct character c after error correction. c It will be the input character 'c' i The probability of the object to be corrected, P(c c P(c) represents the probability of the corrected character appearing in a normal recognition task. i Input character c i The probability of occurrence; frequency(c) c ) represents the character 'c' in the complete identifier encoding library. c The frequency of occurrence, count(c) represents the total number of characters in the complete identifier encoding library; frequency(c) represents the number of occurrences. i |c c ) represents the c in the data where the historical error identification edit distance is 1. c The input character is c. i The number of times the error correction object is defined, where error(c) represents the total amount of historical error identification data with an edit distance of 1.

[0030] A computer system is characterized by comprising: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the method described above.

[0031] A computer-readable storage medium is characterized by storing computer-executable instructions, which, when executed, are used to implement the above-described method.

[0032] A computer program product is characterized by including computer-executable instructions, which, when executed, are used to implement the above-described method.

[0033] The beneficial effects of this invention are as follows:

[0034] This invention provides a post-processing review and correction method for aero-engine blade identification code reading results. First, the blade identification code reading result is input. Based on the Trie data structure and referencing an existing sample library containing all blade identification codes, the accuracy of the reading result is judged using a post-processing review method. If correct, no post-processing correction is performed; if incorrect, the input blade identification code reading result is corrected according to the post-processing correction method provided by this invention, and an interactive selection is performed to output the final correct reading result. Through the application of this invention, the review and correction of aero-engine blade identification code reading results can be realized, achieving a high reading accuracy rate and avoiding a large-scale manual search and verification process, providing a foundation for the informatization and intelligent upgrading of the blade's entire life cycle management process.

[0035] Compared with existing technologies, it has the following technical advantages:

[0036] (1) This invention achieves efficient post-processing review of blade identification coding by introducing a Trie structure, which can be well adapted to scenarios with large amounts of data and improves the speed of the post-processing review process.

[0037] (2) This invention proposes a Bayesian probability calculation method for post-processing correction of encoding, which incorporates historical data experience and effectively solves the problem of the probability priority of candidate results that cannot be effectively distinguished when the edit distance is 1, thereby improving the correction accuracy of the post-processing correction process.

[0038] (3) Based on similarity, edit distance and Bayesian probability value, this invention proposes a sorting scheme for candidate results of error identification coding post-processing correction, which efficiently completes the priority sorting of post-processing correction push results and provides a solution for post-processing correction of error identification results of aero-engine blade identification coding. Attached Figure Description

[0039] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts.

[0040] Figure 1 This is a flowchart of the post-processing review and correction method for the identification and recognition results of aero-engine blade identification codes according to the present invention.

[0041] Figure 2 This is an example of an aircraft engine blade identification code that can be read. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0043] This invention discloses a post-processing review and correction method for the identification results of aero-engine blade identification codes. While machine vision can complete the identification task of aero-engine blade identification codes, it cannot guarantee the absolute accuracy of the identification results. Utilizing the characteristics of the Trie structure, a Trie structure with a complete coding sample library is constructed to achieve efficient post-processing review of the identification results and determine their accuracy. For incorrect identification results, post-processing correction is performed. Based on the idea of ​​dynamic programming, the similarity between the incorrectly identified codes and the codes in the sample library is calculated to obtain a post-processing correction candidate set. Furthermore, dynamic programming is used to calculate the edit distance between the incorrectly identified codes and the post-processing candidate set, proposing a Bayesian probability calculation method when the edit distance is 1. A probability ranking method for the post-processing candidate set based on similarity, edit distance, and Bayesian probability values ​​is given, and the code with the highest probability is selected to complete the post-processing correction of the incorrect identification results.

[0044] like Figure 1 As shown, the specific steps include:

[0045] Step 1.1: Construct the Trie structure for the complete encoded sample library. Read the codes in the encoded sample library one by one. Each code is a key in the Trie structure, and the encoded characters are child nodes of the key. Construct the Trie structure for the complete encoded sample library according to the requirements of the Trie structure.

[0046] Step 1.2: Accuracy assessment of blade encoding recognition results based on the Trie structure. The characters to be reviewed are read sequentially, and it is determined whether each character exists in a child node of the Trie structure in the complete encoding sample library. If it does not exist, it is considered an incorrect recognition; if it exists, the process moves to the next Trie child node to judge the next encoded character. This continues until the last character of the encoding to be reviewed has been judged. If all characters of the encoding to be reviewed exist in the Trie child nodes, the recognition is considered correct.

[0047] In this example, the image being read is as follows: Figure 2As shown, the theoretically correct reading result is {S10.12.3012A,1HZ11211158}, but it was incorrectly read as {S10.12.3012A,1H711211158}. The post-processing review method in step 1 accurately determined the accuracy of the reading result. The complete sample code library used in the post-processing review contains 2000 data entries. The review time for this invention was 6.3249 × 10⁻⁶. -6 The traditional method of sequentially comparing and matching to determine accuracy takes 1.6639 × 10 seconds. -2 The post-processing review method of this invention is more efficient than using seconds.

[0048] Step 2: Post-process and correct the incorrect reading results of the blade identification code, and output the corrected blade identification code reading results to improve the overall reading accuracy.

[0049] Furthermore: Step 2 includes the following steps:

[0050] Step 2.1: Calculate the similarity between the error code recognition result and the sample database code. First, a dynamic programming method is used to find the longest common subsequence between them. Let the error code recognition result X... m Encoding Y with a single sample library n They are respectively "x1x2…x m " and "y1y2...y n "X" m and Y n The longest common subsequence is Z k , denoted as "z1z2...z k LCS(X) m ,Y n ) represents X m and Y n The longest common subsequence can be found by the following conclusion:

[0051]

[0052] The longest common subsequence problem exhibits the characteristics of optimal substructure in dynamic programming. The optimal solutions to its subproblems can help solve the optimal solution to the original problem. The optimal solution to the original problem can be obtained by combining the optimal solutions to the subproblems, as shown in the following conclusion:

[0053]

[0054] In the formula C(X) m ,Y n ) represents X m and Y n The number of elements in the longest common subsequence of sequence X, where m and n represent the sequence X. m and Y nThe number of elements, where m and n are 0, represents X. m and Y n The sequence is empty. Error-coded reading result X m Encoding Y with a single sample library n The similarity between them is R, and the formula is as follows:

[0055]

[0056] in, and Representing sequence X m and Y n The number of elements.

[0057] The similarity between the error code recognition result and a single code in the sample database is calculated one by one according to the formula, resulting in a list of similarity values.

[0058] Step 2.2: Generate a post-processing correction candidate set for the misread results. Sort the similarity list obtained in Step 2.1 according to the similarity value, and select the top n sample database codes to form the post-processing correction candidate set according to the actual task requirements.

[0059] In this example, the top 5 codes with the highest similarity values ​​are selected to form the post-processing correction candidate set.

[0060] Step 2.3: Calculate the edit distance between the error identification result and the post-processing correction candidate set. Analyze the error identification results; all errors can be categorized into three types: replacement, insertion, and deletion. Replacement, insertion, and deletion are all considered as one modification. The error code reading result X is then... m Modified to a single sample library encoding Y n The minimum number of edits required is called the edit distance. The problem of finding the edit distance also exhibits the optimal substructure characteristic of dynamic programming and can be solved using dynamic programming methods. lev(X) m ,Y n ) represents X m and Y n The edit distance satisfies the following conclusion:

[0061]

[0062] The edit distance between the error code reading result and the post-processing correction candidate set is calculated one by one according to the formula, resulting in a list of edit distance values.

[0063] Step 2.4: Calculate the Bayesian probability when the edit distance is 1. Without considering other interfering factors, the probabilities of replacement, insertion, and deletion errors are the same, and the edit distance is 1 for all three. However, the encoding similarity of the three error identifications differs significantly, making it impossible to rank their probabilities. This invention addresses this problem by calculating the Bayesian probability of candidate codes when the edit distance is 1 based on a Bayesian algorithm. First, character-level error localization is performed, denoted as c. i and c c These represent the input character and the corrected character in the character-level problem localization result, respectively. For ease of representation, the input character in the insertion error type correction is defined as "insert," and the corrected character in the deletion error type correction is defined as "delete." Through character-level problem localization, the post-encoding correction problem with an edit distance of 1 can be transformed into a character-level error correction problem, achieving problem downgrading. The Bayesian calculation formula is shown below:

[0064]

[0065] In the formula P(c c |c i ) indicates that the input character is c i After correction, the character is 'c'. c The probability, P(c) i |c c ) indicates the correct character c after error correction. c It will be the input character 'c' i The probability of the object to be corrected, P(c c P(c) represents the probability of the corrected character appearing in a normal recognition task. i Input character c i The probability of occurrence. P(c) c The calculation can be performed based on historical dataset information, and the formula is shown below:

[0066]

[0067] In the formula, frequency(c c ) represents the character 'c' in the complete identifier encoding library. c The number of occurrences, count(c) represents the total number of characters in the complete identifier encoding library. P(c) i |c c The calculation formula for ) is as follows:

[0068]

[0069] In the formula, frequency(c i |c c) represents the c in the data where the historical error identification edit distance is 1. c The input character is c. i The number of times the error correction object is defined, where error(c) represents the total number of historical error identification data with an edit distance of 1. In summary, P(c) can be achieved. c |c i The calculation of P(c) is performed. All P(c) in the candidate set with an edit distance of 1 are considered. c |c i By sorting the values ​​from largest to smallest, the best candidate correction identifier code can be obtained.

[0070] Step 2.5: Perform a comprehensive ranking of candidate results. Based on the data obtained in the above steps, the post-processing correction candidate set is first divided into two categories according to the edit distance: edit distance 1 and the other two. The category with edit distance 1 is ranked according to Bayesian probability, and the other categories are directly ranked according to similarity. The category with edit distance 1 has higher priority than the other categories. The probability ranking of the candidate results is obtained by combining the results, and the candidate code with the highest probability at the top is output as the correction result.

[0071] In this example, the similarity, edit distance, and Bayesian probability calculation results during the error identification coding post-processing correction are shown in the table below:

[0072]

[0073]

[0074] Following the method in step 2, the post-processing correction result for this example is {S10.12.3012A,1HZ11211158}, which is consistent with the theoretically correct reading result.

[0075] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the scope of the technology disclosed in the present invention, and such modifications or substitutions should all be covered within the scope of protection of the present invention.

Claims

1. A method for post-processing, reviewing, and correcting the identification results of aero-engine blade identification codes, characterized in that, The steps are as follows: Step 1: Utilizing the characteristics of the Trie structure, construct a complete Trie structure for the encoded sample library, and perform post-processing review on the leaf identification encoding recognition results to determine the accuracy of the recognition results; Step 1 includes: Step 1.1: Construct the Trie structure of the complete encoded sample library. Read the codes in the encoded sample library one by one. Each code is a key of the Trie structure, and the characters of the code are child nodes of the key. Construct the Trie structure of the complete encoded sample library according to the requirements of the Trie structure. Step 1.2: Accuracy judgment of blade code reading results based on Trie structure: Read the code character to be examined in sequence and judge whether the character exists in the child node of the Trie structure of the complete code sample library. If it does not exist, it is an incorrect reading. If it exists, move to the next Trie child node to judge the next code character until the last character of the code to be examined is judged. If all characters of the code to be examined exist in the Trie child node, the reading is correct. Step 2: Post-processing correction is performed on the coding error identification results. Based on the idea of ​​dynamic programming, the similarity between the error identification code and the sample library code is calculated to obtain a post-processing correction candidate set. Dynamic programming is then used to solve for the edit distance between the error identification code and the post-processing candidate set. The Bayesian probability when the edit distance is 1 is calculated. Based on the similarity, edit distance, and Bayesian probability values, a probability ranking method for the post-processing candidate set is used to select the code with the highest probability to complete the post-processing correction of the error identification results. Step 2 includes: Step 2.1: Calculate the similarity between the error code recognition result and the sample library code, and obtain a list of similarity values; Step 2.2: Sort the similarity list according to the size of the similarity value, and select the top n sample database codes to form a post-processing correction candidate set according to the actual task requirements; Step 2.3: Calculate the edit distance between the error identification result and the post-processing correction candidate set to obtain a list of edit distance values; Step 2.4: Calculate the Bayesian probability when the edit distance is 1, and select all candidates with an edit distance of 1 from the candidate set. By sorting the values ​​from largest to smallest, the best candidate correction identifier code can be obtained; Step 2.5: Comprehensive ranking of candidate results: Based on the data obtained in the above steps, the post-processing correction candidate set is first divided into two categories according to edit distance: edit distance 1 and others. The category with edit distance 1 is ranked according to Bayesian probability, and the other categories are directly ranked according to similarity. The category with edit distance 1 has higher priority than the other categories. The probability ranking of the candidate results is obtained by comprehensive ranking, and the candidate code with the highest probability at the top is output as the correction result; the calculation of the Bayesian probability when the edit distance is 1 is specifically as follows: in, This indicates that when the input character is After correction, the character is The probability, This indicates the corrected characters after error correction. It will be the input characters The probability of the error correction target. This indicates the probability of the corrected character appearing in a normal recognition task. Input characters The probability of occurrence; Represents characters in the complete identifier encoding library Number of times it appears This represents the total number of characters in the complete identifier encoding library; This indicates that in the data where the historical error identification edit distance is 1 It is the input character The number of times the error correction object is performed. This represents the total amount of historical error identification data with an edit distance of 1.

2. The method for post-processing review and correction of the identification results of aero-engine blade identification codes according to claim 1, characterized in that, The calculation of the similarity between the error code recognition result and the sample library code is specifically as follows: in, express and The number of elements in the longest common subsequence. , Representative sequence and The number of elements, when , When it is 0, it represents and An empty sequence and Representative sequence and The number of elements.

3. The method for post-processing review and correction of the identification results of aero-engine blade identification codes according to claim 1, characterized in that, The calculation of the edit distance between the error identification result and the post-processing correction candidate set is specifically as follows: in, express and Edit distance.

4. A computer system, characterized in that... include: One or more processors, a computer-readable storage medium, for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to perform the method of any one of claims 1-3.

5. A computer-readable storage medium, characterized in that... The device stores computer-executable instructions, which, when executed, are used to implement the method described in any one of claims 1-3.

6. A computer program product, characterized in that... It includes computer-executable instructions, which, when executed, are used to implement the method of any one of claims 1-3.

Images