Artificial intelligence speech recognition method based on industrial Internet

Abstract

The present invention relates to the field of intelligent speech recognition, and in particular to an artificial intelligence speech recognition method based on the industrial Internet. A speech receiver cluster is set. When a recognition device detects that a product to be detected has passed, a trigger instruction is issued to perform a quality test. The speech receiver pre-learns the quality test audio corresponding to each product to be detected to form audio pre-learning data. An audio learning model generates a quality test distribution map, which is compared with a quality test threshold to form a quality test result. The position of the trigger element is adjusted again according to the quality test result, and the quality test result of the product to be detected is obtained again. Faulty products that have not passed the quality test are screened, a focusing instruction is issued to the recognition device, and a high-definition image of the faulty product is taken and uploaded to a recognition platform. Through speech recognition and machine learning, automated quality inspection of assembly line products is realized, and the inspection strategy can be adjusted in real time to adapt to different products and quality requirements, thereby improving inspection efficiency and product quality.

Description

Technical Field

[0001] The present invention relates to the field of intelligent speech recognition, and in particular to an artificial intelligence speech recognition method based on the industrial Internet. Background Art

[0002] The development and application of deep learning technology have promoted the advancement of intelligent speech recognition technology and expanded its application scope. Currently, industrial-grade speech recognition systems usually use deep learning technologies, such as convolutional neural networks (CNN) and recurrent neural networks (RNN). These technologies can effectively extract the features of speech signals and improve recognition accuracy.

[0003] Due to its complex, diverse and highly professional characteristics, the industrial field lacks excellent industrial theme AI data models and industrial annotated data sets for AI algorithm training. In addition, the construction of AI infrastructure in the industrial field is relatively backward, which restricts the development of industrial intelligence.

[0004] Chinese patent application publication number: CN108492822A discloses a voice recognition method based on industrial applications, including the following steps: (1) sending a specific task message to the device end through the server end; (2) the device end converts the message, performs signal noise reduction processing, and plays it out in the form of voice; after the receiving unit of the device end receives the message, the control unit will output the control instruction to the broadcast unit, and the received message will be broadcasted by voice through the broadcast unit; (3) the device end recognizes, processes and converts the voice through the voice recognition module, and converts the voice into text or command execution operation through voice analysis; the device end sends the voice instruction through the voice control interface, recognizes the voice instruction through the receiving module, and the output unit executes the instruction. The method of the invention effectively reduces the workload of the staff in the industrial operation process and improves the work efficiency of the operating personnel when operating complex systems.

[0005] Chinese patent application publication number: CN113470652A discloses a speech recognition and processing method based on the industrial Internet, which relates to the field of speech recognition technology. The invention includes the following steps: constructing a text database corresponding to the industrial Internet; obtaining speech data to be recognized; preprocessing the speech data to be recognized; extracting features from the processed speech data; matching and comparing the speech features to be recognized with the acoustic model to obtain the recognition result; inputting the recognition result into the language model for language processing to obtain grammatical and semantic analysis results. The invention can improve the accuracy of speech recognition and eliminate the interference caused by noise by constructing an industrial Internet text database, preprocessing and extracting features from the speech data to be recognized, matching the extracted features with the acoustic model, obtaining the recognition results and inputting them into the speech model for analysis to obtain grammatical and semantic analysis results.

[0006] However, the above method has the following problems: it is impossible to use intelligent voice recognition technology to detect product quality, thereby failing to improve the industrial production process. Summary of the invention

[0007] To this end, the present invention provides an artificial intelligence speech recognition method based on the industrial Internet to overcome the problem in the prior art that intelligent speech recognition technology cannot be used to detect product quality, thereby failing to improve the industrial production process.

[0008] To achieve the above object, the present invention provides an artificial intelligence speech recognition method based on industrial Internet, comprising:

[0009] Set up a voice receiver cluster starting from the product entrance of the production line to be tested;

[0010] When the product to be inspected passes through the identification device, a trigger instruction is issued to the trigger element to control the trigger element to adjust its position and perform a quality test on the product to be inspected;

[0011] The voice receiver receives and pre-learns the quality test audio corresponding to each product to be tested to form corresponding audio pre-learning data;

[0012] The audio learning model learns the audio pre-learning data to generate a corresponding quality test distribution graph;

[0013] Comparing the quality test distribution diagram with the quality test threshold to form a corresponding quality test result, adjusting the position of the trigger element again according to the quality test result, and obtaining the quality test result of the product to be tested again;

[0014] Screen out faulty products that fail the quality test, respond to the trigger instruction and issue a focusing instruction to the recognition device, take a high-definition image of the faulty product and upload it to the recognition platform.

[0015] Further, the step of forming a speech receiver cluster includes:

[0016] determining the geometric center of the product inlet;

[0017] Taking the geometric center as a reference, a number of voice receivers are arranged in the pipeline to be detected at random integer multiples of a preset spacing;

[0018] connecting the voice receiver to the recognition platform to form a voice receiver cluster;

[0019] Wherein, the voice receiver and the recognition platform establish a connection via the mobile Internet.

[0020] Furthermore, the step of issuing a trigger instruction to the trigger element includes:

[0021] When the recognition device recognizes that the product to be detected passes through, it takes a simple image of the product to be detected;

[0022] Determine the product information of the product to be detected according to the simplified image, wherein:

[0023] The product information includes the size and / or material of the product to be tested;

[0024] Determine the position parameter of the trigger element according to the product information, and issue a trigger instruction to the trigger element, wherein:

[0025] The trigger instruction includes the product information and the location parameter.

[0026] Furthermore, the steps of performing quality testing on the product to be tested include:

[0027] The trigger element receives the trigger instruction and adjusts to the position corresponding to the position parameter;

[0028] The trigger element drives the knocking head to knock the surface of the product to be tested to generate corresponding quality test audio;

[0029] Wherein, the trigger element is connected to the striking head via a sliding rod;

[0030] The trigger element drives the striking head by controlling the sliding rod to move repeatedly;

[0031] The knocking force of the knocking head on each product to be tested is the same.

[0032] Furthermore, the step of forming audio pre-learning data includes:

[0033] The voice receiver marks the quality test audio in a test order;

[0034] Select several index features of the quality test audio, where:

[0035] The indicator features include the frequency, volume and / or duration of the sound;

[0036] According to the test sequence, the quality test audio is segmented according to the standard learning rate to form corresponding audio pre-learning data, wherein:

[0037] The standard learning rate is a learning rate that can be recognized by the audio learning model, and for a single learning, the corresponding standard learning rate is a single learning rate.

[0038] Furthermore, the step of generating a quality test distribution diagram includes:

[0039] Passing the audio pre-learning data and the indicator features into the audio learning model;

[0040] The audio learning model learns the audio pre-learning data to generate a corresponding quality test distribution graph;

[0041] The quality test distribution diagram is labeled according to the indicator characteristics.

[0042] Further, the steps of forming the quality test results include:

[0043] comparing the quality test distribution map with the quality test threshold;

[0044] Marking the qualified status of the product to be tested;

[0045] Calculate the average qualified rate of the products to be tested, and compare the average qualified rate with the qualified rate threshold to form a corresponding quality test result;

[0046] Wherein, when the value of the quality test distribution graph is greater than the quality test threshold, the corresponding product to be tested is marked as a qualified product, and when the value of the quality test distribution graph is less than the quality test threshold, the corresponding product to be tested is marked as a faulty product;

[0047] The quality test threshold is related to the preset distance and / or the number of the voice receivers;

[0048] The average qualified rate is the proportion of qualified products in all the products to be tested;

[0049] The pass rate threshold is the minimum standard of the pass rate corresponding to the quality test.

[0050] Furthermore, when the average qualified rate is greater than the qualified rate threshold, the product to be inspected is shipped out of the factory, and when the average qualified rate is less than the qualified rate threshold, the position of the trigger element is adjusted accordingly.

[0051] Furthermore, the product to be tested is quality tested again, and the corresponding secondary qualified rate is counted. When the secondary qualified rate is greater than the qualified rate threshold, the corresponding test result is marked as passed, and the product to be tested is shipped out of the factory.

[0052] Furthermore, when the secondary pass rate is less than the pass rate threshold, the corresponding test result is marked as failed, and the voice receiver responds to the trigger instruction to issue a focusing instruction to the recognition device, determines the object distance between the recognition device and the product to be inspected, and focuses the recognition device based on a preset mapping relationship between the focus position and the object distance, and takes a high-definition image corresponding to the faulty product in the quality test distribution map, and uploads it to the recognition platform.

[0053] Compared with the prior art, the present invention sets up a voice receiver cluster. When the recognition device detects that the product to be detected has passed, it issues a trigger instruction to perform a quality test. The voice receiver pre-learns the quality test audio corresponding to each product to be detected to form audio pre-learning data. The audio learning model generates a quality test distribution map, which is compared with the quality test threshold to form a quality test result. The position of the trigger element is adjusted again according to the quality test result, and the quality test result of the product to be detected is obtained again. Faulty products that have not passed the quality test are screened, and a focusing instruction is issued to the recognition device. High-definition images of the faulty products are taken and uploaded to the recognition platform. Through voice recognition and machine learning, automated quality inspection of assembly line products is realized, and the inspection strategy can be adjusted in real time to adapt to different products and quality requirements, thereby improving inspection efficiency and product quality.

[0054] Furthermore, by setting up a voice receiver cluster, real-time data transmission is ensured, so that quality inspection results can be quickly fed back and the production process can be adjusted in time, which improves the operability and convenience of the system and makes the quality inspection of the assembly line more automated and intelligent.

[0055] Furthermore, by taking a simple image of the product to be inspected, the product information of the product to be inspected is determined, and a trigger instruction is issued to the trigger element, the quality test can be adapted to products of different sizes and materials. By identifying the specific characteristics of the product and adjusting the position of the trigger element accordingly, the accuracy of the quality test is improved, and the flexibility and applicability of the test are enhanced.

[0056] Furthermore, by collecting quality test audio of the products to be tested and performing quality tests on the products to be tested, the same knocking force is ensured, making the test results between different products comparable. By precisely controlling the knocking position and force, the accuracy of the test results is improved, manual operations are reduced, and the efficiency and reliability of quality testing are improved.

[0057] Furthermore, by marking the test order of the quality test audio, selecting several indicator features of the quality test audio, and segmenting the quality test audio according to the test order at a standard learning rate, corresponding audio pre-learning data is formed, thereby achieving orderly labeling and segmentation of audio data, and systematically constructing a pre-learning data set, which helps the model capture quality-related signals of the product and enables the model to learn from audio data more intelligently and predict product quality.

[0058] Furthermore, by comparing the quality test distribution map with the quality test threshold, the qualified status of the products to be tested is marked, and the average qualified rate of the products to be tested is calculated to form the corresponding quality test results. The position of the trigger element is adjusted again, and the corresponding secondary qualified rate is generated. By comparing the test results with the threshold in real time, a judgment on whether the product is qualified can be made quickly, and the test parameters can be adjusted as needed to adapt to different production conditions, thereby effectively controlling the quality of the product and improving the accuracy and efficiency of decision-making.

[0059] Furthermore, a focusing instruction is issued to the recognition device through the voice receiver, and the corresponding faulty products in the quality test distribution map are photographed to form high-definition images and uploaded to the recognition platform. Through high-definition image capture, product defects are accurately identified, providing traceable records for product quality control. Utilizing the intelligent analysis tools of the recognition platform, product quality problems are traced more deeply, continuously improving production processes and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of the artificial intelligence speech recognition method based on the industrial Internet of the present invention;

[0061] Figure 2 A flow chart for forming a speech receiver cluster for the present invention;

[0062] Figure 3 A flow chart of issuing a trigger instruction to a trigger element according to the present invention;

[0063] Figure 4 Flowchart for forming audio pre-learning data for the present invention. DETAILED DESCRIPTION

[0064] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0065] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0066] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0067] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0068] See also Figure 1 As shown, it is a flow chart of the artificial intelligence speech recognition method based on the industrial Internet of the present invention, including:

[0069] Step S1, setting a voice receiver cluster starting from the product entrance of the assembly line to be tested;

[0070] Step S2, when the product to be inspected passes through the identification device, a trigger instruction is issued to the trigger element to control the trigger element to adjust its position and perform a quality test on the product to be inspected;

[0071] Step S3, the voice receiver receives and pre-learns the quality test audio corresponding to each product to be tested to form corresponding audio pre-learning data;

[0072] Step S4, the audio learning model learns the audio pre-learning data and generates a corresponding quality test distribution graph;

[0073] Step S5, comparing the quality test distribution map with the quality test threshold to form a corresponding quality test result, adjusting the position of the trigger element again according to the quality test result, and obtaining the quality test result of the product to be tested again;

[0074] Step S6, screening out faulty products that have not passed the quality test, responding to the trigger instruction and issuing a focusing instruction to the recognition device, taking a high-definition image of the faulty product and uploading it to the recognition platform.

[0075] By setting up a voice receiver cluster, when the recognition device detects that the product to be inspected has passed, a trigger command is issued to perform a quality test. The voice receiver pre-learns the quality test audio corresponding to each product to be inspected to form audio pre-learning data. The audio learning model generates a quality test distribution map, which is compared with the quality test threshold to form a quality test result. The position of the trigger element is adjusted again according to the quality test result, and the quality test result of the product to be inspected is obtained again to screen out faulty products that have not passed the quality test. A focusing command is issued to the recognition device, and high-definition images of the faulty products are taken and uploaded to the recognition platform. Through voice recognition and machine learning, automated quality inspection of assembly line products is realized, and the inspection strategy can be adjusted in real time to adapt to different products and quality requirements, thereby improving inspection efficiency and product quality.

[0076] See also Figure 2 As shown, it is a flow chart of forming a speech receiver cluster according to the present invention, including:

[0077] Step S11, determining the geometric center of the product entrance;

[0078] Step S12, taking the geometric center as a reference, setting a number of voice receivers in the pipeline to be detected at random integer multiples of a preset spacing;

[0079] Step S13, connecting the voice receiver to the recognition platform to form a voice receiver cluster;

[0080] Among them, the voice receiver and the recognition platform establish a connection through the mobile Internet.

[0081] In the specific implementation, the geometric center of the pipeline product entrance is determined. This center point will be used as the reference point for laying out the voice receiver cluster to ensure that the receivers are evenly laid out and cover the entire pipeline.

[0082] Based on the geometric center, several voice receivers are set in the inspection line according to the preset spacing (this spacing can be fixed or a random integer multiple to increase the flexibility and adaptability of the layout). Such a layout can ensure that each receiver can effectively capture the audio signal generated when the product passes.

[0083] The connection between the voice receiver and the recognition platform is established through the mobile Internet. This connection method provides flexibility and remote access capabilities, and realizes real-time data transmission and processing. Real-time data transmission allows quality inspection results to be quickly fed back to the production process, allowing timely adjustments to optimize production efficiency and product quality.

[0084] By setting up a voice receiver cluster, real-time data transmission is ensured, so that quality inspection results can be quickly fed back and the production process can be adjusted in time, which improves the operability and convenience of the system and makes the quality inspection of the assembly line more automated and intelligent.

[0085] See also Figure 3 As shown, it is a flow chart of the present invention issuing a trigger instruction to a trigger element, including:

[0086] Step St1, when the recognition device recognizes that the product to be detected passes through, it takes a simple image of the product to be detected;

[0087] Step St2, determining the product information of the product to be detected based on the simplified image, wherein:

[0088] Product information includes the size and / or material of the product to be tested;

[0089] Step St3, determining the position parameters of the trigger element according to the product information, and issuing a trigger instruction to the trigger element, wherein:

[0090] The trigger instruction includes product information and location parameters.

[0091] In the specific implementation, when the recognition device recognizes that the product to be inspected has passed, it will take a simple image of the product to obtain preliminary visual information of the product, which provides a basis for subsequent product information recognition. Based on the determined product information, the system will determine the position parameters of the trigger element, and accordingly issue a trigger instruction containing product information and position parameters to the trigger element, ensuring that the trigger element can be accurately adjusted according to the specific characteristics of the product to adapt to products of different sizes and materials.

[0092] By taking simple images and determining product information based on them, the system can more accurately control the position of the trigger element, thereby improving the accuracy of quality testing.

[0093] This process can adapt to products of different sizes and materials. It adjusts the position of the trigger element by identifying the specific characteristics of the product, thereby improving the applicability of quality testing. The position of the trigger element is dynamically adjusted according to the real-time product information to adapt to the ever-changing product characteristics, making the quality inspection of the assembly line more automated and intelligent.

[0094] By taking a simple image of the product to be tested, determining the product information of the product to be tested, and issuing a trigger instruction to the trigger element, the quality test can be adapted to products of different sizes and materials. By identifying the specific characteristics of the product and adjusting the position of the trigger element accordingly, the accuracy of the quality test is improved, and the flexibility and applicability of the test are enhanced.

[0095] Specifically, the steps for quality testing of the products to be tested include:

[0096] The trigger element receives the trigger instruction and adjusts to a position corresponding to the position parameter;

[0097] The trigger element drives the knocking head to knock on the surface of the product to be tested, generating the corresponding quality test audio;

[0098] Wherein, the trigger element is connected to the striking head via a sliding rod;

[0099] The trigger element drives the striking head by controlling the sliding rod to move repeatedly;

[0100] The knocking force of the knocking head on each product to be tested is the same.

[0101] In a specific implementation, the trigger element receives a trigger instruction from the identification device and adjusts to a corresponding position according to the position parameters, ensuring that the trigger element can accurately align with a specific part of the product to be tested. The trigger element drives the knocking head to knock on the surface of the product to be tested, generates a corresponding quality test audio, and evaluates the quality of the product through audio analysis. The trigger element is connected to the knocking head through a sliding rod, and the trigger element drives the knocking head by controlling the repeated movement of the sliding rod. The design of this mechanical structure allows precise control of the movement of the knocking head to ensure the consistency and repeatability of the knocking. Since the knocking force is the same, the test results between different products can be directly compared, which helps to identify quality differences. By precisely controlling the knocking position and force, the accuracy of the test results is improved because each test is performed under strictly controlled conditions. By reducing manual intervention and improving the consistency of testing, the test results are more reliable, which helps to more accurately evaluate the quality of the product.

[0102] By collecting the quality test audio of the product to be tested and performing quality tests on the product to be tested, the same knocking force is ensured, making the test results between different products comparable. By precisely controlling the knocking position and strength, the accuracy of the test results is improved, manual operations are reduced, and the efficiency and reliability of quality testing are improved.

[0103] See also Figure 4 As shown, it is a flow chart of the present invention for forming audio pre-learning data, including:

[0104] Step Sp1, the voice receiver marks the quality test audio in a test order;

[0105] Step Sp2, select several index features of the quality test audio, where:

[0106] Indicator characteristics include the frequency of the sound, the volume, and / or the duration of the sound;

[0107] Step Sp3, according to the test sequence, the quality test audio is segmented according to the standard learning rate to form corresponding audio pre-learning data, wherein,

[0108] The standard learning rate is a learning rate that the audio learning model can recognize, and for single-shot learning, the corresponding standard learning rate is a single learning rate.

[0109] In some possible implementations, when the standard learning rate is set at 160 / second to 180 / second, the test result of the quality test is most accurate. Generally, the standard learning rate is set to 170 / second.

[0110] In a specific implementation, the voice receiver marks the collected quality test audio in the order of the test. This step helps to maintain the order and time correlation of the data in subsequent analysis, because the order of the audio data is closely related to the corresponding product quality characteristics. Several indicator features are selected from the quality test audio, including the frequency of the sound, the volume and / or the duration of the sound. These features can reflect the physical response of the product when it is knocked, and are an important basis for judging the quality of the product. According to the test order, the quality test audio is segmented according to the standard learning rate to form the corresponding audio pre-learning data. The standard learning rate refers to the learning rate that the audio learning model can recognize and process. For a single learning, the corresponding standard learning rate is a single learning rate. This step ensures that the audio data can be effectively learned and processed by the model. By orderly marking and segmenting the audio data, a pre-learning data set can be systematically constructed. The selected key indicator features help the model capture signals related to product quality, so that the model can extract useful information from the audio data.

[0111] By marking the test order of the quality test audio, selecting several indicator features of the quality test audio, and segmenting the quality test audio according to the test order at a standard learning rate, corresponding audio pre-learning data is formed. This achieves orderly marking and segmentation of audio data, and systematically constructs a pre-learning data set, which helps the model capture quality-related signals of the product and enables the model to learn from audio data more intelligently and predict product quality.

[0112] Specifically, the steps of generating a quality test distribution map include:

[0113] Passing audio pre-learning data and indicator features into the audio learning model;

[0114] The audio learning model learns the audio pre-learning data and generates a corresponding quality test distribution graph;

[0115] The quality test distribution map is labeled according to the indicator characteristics.

[0116] In the specific implementation, the audio learning model learns and analyzes the incoming audio pre-learning data. In this process, the model will identify and learn patterns and trends in the audio data, which are associated with specific quality characteristics of the product. The model will use various algorithms, such as deep learning, support vector machines, or other classification and regression techniques, to process the audio data. After learning, the audio learning model generates a corresponding quality test distribution map. This distribution map is a visual representation of the quality information contained in the audio data. It includes the distribution of audio features, the intensity of frequency, changes in volume, and the duration of sound. Labeling the quality test distribution map according to the selected indicator features helps to interpret the data points in the distribution map and associate them with specific quality attributes, making it easier to understand and analyze the test results. Using machine learning, the model is able to identify subtle quality differences from complex audio data, improving the accuracy of quality detection.

[0117] Specifically, the steps to form quality test results include:

[0118] comparing the quality test distribution map to a quality test threshold;

[0119] Mark the qualified status of the products to be tested;

[0120] Calculate the average qualified rate of the products to be tested, and compare the average qualified rate with the qualified rate threshold to form the corresponding quality test results;

[0121] When the value of the quality test distribution diagram is greater than the quality test threshold, the corresponding product to be tested is marked as a qualified product; when the value of the quality test distribution diagram is less than the quality test threshold, the corresponding product to be tested is marked as a faulty product;

[0122] The quality test threshold is related to a preset spacing and / or the number of voice receivers;

[0123] The average qualified rate is the proportion of qualified products among all the products to be tested;

[0124] The pass rate threshold is the minimum standard for the pass ratio corresponding to the quality test.

[0125] In some possible implementations, if the quality test threshold is 80, and the value corresponding to product A on the quality test distribution map is 85, which is greater than the quality test threshold, product A is marked as a qualified product. If the value corresponding to product B on the quality test distribution map is 75, which is less than the quality test threshold, product B is marked as a faulty product.

[0126] Specifically, when the average pass rate is greater than the pass rate threshold, the product to be tested is shipped out of the factory, and when the average pass rate is less than the pass rate threshold, the position of the trigger element is adjusted accordingly.

[0127] In some possible implementations, if the pass rate threshold is 80%, there are 100 products to be inspected on the assembly line, 85 are qualified products, and 15 are faulty products, then the average pass rate is 85%, and the average pass rate is greater than the pass rate threshold, and the products to be inspected on the assembly line will be shipped out.

[0128] If the pass rate threshold is 80%, there are 100 products to be tested on the assembly line, 75 are qualified products, and 25 are faulty products, then the average pass rate is 75%, and the average pass rate is less than the pass rate threshold. Correspondingly, the position of the trigger element is adjusted to the center position of the product to be tested, and the product to be tested is quality tested again.

[0129] In the specific implementation, the value of each product to be tested on the quality test distribution map is compared with the preset quality test threshold. This step is the basis for judging whether the product is qualified. Through specific value comparison and threshold setting, product quality can be accurately controlled to ensure that only products that meet the standard quality can be shipped. When the average pass rate is lower than the pass rate threshold, the production process can be adjusted in time to reduce the production of unqualified products. Automated quality testing and decision-making processes can improve production efficiency and reduce the cost and time of manual inspection. All decisions are based on actual test data, making decisions more objective and reliable.

[0130] Specifically, the product to be tested is quality tested again, and the corresponding secondary qualified rate is counted. When the secondary qualified rate is greater than the qualified rate threshold, the corresponding test result is marked as passed, and the product to be tested is shipped out of the factory.

[0131] In some possible implementations, if the pass rate threshold is 80%, there are 100 products to be inspected on the assembly line, and the products to be inspected are quality tested again, the results show that there are 85 qualified products and 15 faulty products, then the secondary pass rate is 85%, and the secondary pass rate is greater than the pass rate threshold, and the products to be inspected on the assembly line are shipped out.

[0132] Specifically, when the secondary pass rate is less than the pass rate threshold, the corresponding test result will be marked as failed. The voice receiver responds to the trigger command and issues a focusing command to the recognition device to determine the object distance between the recognition device and the product to be tested. The recognition device is focused based on the mapping relationship between the preset focus position and the object distance, and a high-definition image corresponding to the faulty product in the quality test distribution map is taken and uploaded to the recognition platform.

[0133] In some possible implementations, if the pass rate threshold is 80%, there are 100 products to be inspected on the assembly line, and the products to be inspected are quality tested again, the results show that there are 75 qualified products and 25 faulty products, then the secondary pass rate is 75%, and the secondary pass rate is less than the pass rate threshold. The voice receiver responds to the trigger command and issues a focusing command to the recognition device.

[0134] In the specific implementation, after receiving the trigger command, the voice receiver will issue a focusing command to the recognition device (such as a camera or visual sensor). This command is to prepare to take a high-definition image of the faulty product. The recognition device needs to determine the object distance to the product to be detected, that is, the actual distance between the recognition device and the product to be detected. According to the mapping relationship between the preset focus position and the object distance, the recognition device automatically adjusts the focal length for focusing. The mapping relationship refers to the corresponding focal length that the recognition device should be adjusted to at different object distances to ensure a clear image. Once the focusing is completed, the recognition device takes high-definition images corresponding to the faulty product. These images will contain detailed visual information of the product, which will help subsequent fault analysis and quality control. The captured high-definition images are uploaded to the recognition platform, which can be a central server or cloud platform for storing and analyzing image data. High-definition images are used for subsequent fault diagnosis to help engineers understand the cause of the fault and take corresponding corrective measures, providing support for product fault analysis and continuous improvement.

[0135] By comparing the quality test distribution map with the quality test threshold, the qualified status of the products to be tested is marked, and the average qualified rate of the products to be tested is calculated to form the corresponding quality test results. The position of the trigger element is adjusted again, and the corresponding secondary qualified rate is generated. By comparing the test results with the threshold in real time, a judgment on whether the product is qualified can be made quickly, and the test parameters can be adjusted as needed to adapt to different production conditions, effectively controlling the quality of the product and improving the accuracy and efficiency of decision-making.

[0136] A focusing command is issued to the recognition device through the voice receiver, and the corresponding faulty products in the quality test distribution map are photographed to form high-definition images and uploaded to the recognition platform. Through high-definition image capture, product defects are accurately identified, providing traceable records for product quality control. The intelligent analysis tools of the recognition platform are used to trace product quality problems more deeply, continuously improving production processes and product quality.

[0137] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0138] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An artificial intelligence speech recognition method based on industrial Internet, characterized in that: include: Set up a voice receiver cluster starting from the product entrance of the production line to be tested; When the product to be inspected passes through the identification device, a trigger instruction is issued to the trigger element to control the trigger element to adjust its position and perform a quality test on the product to be inspected; Each voice receiver receives and pre-learns the quality test audio corresponding to each product to be tested, forming corresponding audio pre-learning data; The audio learning model learns the audio pre-learning data to generate a corresponding quality test distribution graph; Comparing the quality test distribution diagram with a quality test threshold to form a corresponding quality test result, adjusting the position of the trigger element again according to the quality test result, and obtaining the quality test result of the product to be tested again; Screening out faulty products that have not passed the quality test, responding to the trigger instruction and issuing a focusing instruction to the recognition device, taking a high-definition image of the faulty product and uploading it to the recognition platform; The steps of performing the quality test on the product to be tested include: The trigger element receives the trigger instruction and adjusts to a position corresponding to the position parameter; The trigger element drives the knocking head to knock the surface of the product to be tested to generate corresponding quality test audio; Wherein, the trigger element is connected to the striking head via a sliding rod; The trigger element drives the striking head by controlling the sliding rod to move repeatedly; The knocking force exerted by the knocking head on each product to be tested is the same; The steps of forming the audio pre-learning data include: The voice receiver marks the quality test audio in a test order; Select several index features of the quality test audio, where: The indicator features include the frequency, volume and / or duration of the sound; According to the test sequence, the quality test audio is segmented according to the standard learning rate to form corresponding audio pre-learning data, wherein: The standard learning rate is a learning rate that can be recognized by the audio learning model, and for a single learning, the corresponding standard learning rate is a single learning rate; The steps of generating the quality test distribution diagram include: Passing the audio pre-learning data and the indicator features into the audio learning model; The audio learning model learns the audio pre-learning data to generate a corresponding quality test distribution graph; Marking the quality test distribution map according to the indicator characteristics; When the value of the quality test distribution diagram is greater than the quality test threshold, the corresponding product to be tested is marked as a qualified product. When the value of the quality test distribution diagram is less than the quality test threshold, the corresponding product to be tested is marked as a faulty product.

2. The artificial intelligence speech recognition method based on industrial Internet according to claim 1 is characterized in that: The steps to form a speech receiver cluster include: determining the geometric center of the product inlet; Taking the geometric center as a reference, a number of voice receivers are arranged in the pipeline to be detected at random integer multiples of a preset spacing; connecting the voice receiver to the recognition platform to form a voice receiver cluster; Wherein, the voice receiver and the recognition platform establish a connection via the mobile Internet.

3. The artificial intelligence speech recognition method based on industrial Internet according to claim 2 is characterized in that: The step of issuing a trigger instruction to the trigger element comprises: When the recognition device recognizes that the product to be detected passes through, it takes a simple image of the product to be detected; Determine the product information of the product to be detected according to the simplified image, wherein: The product information includes the size and / or material of the product to be tested; Determine the position parameter of the trigger element according to the product information, and issue a trigger instruction to the trigger element, wherein: The trigger instruction includes the product information and the location parameter.

4. The artificial intelligence speech recognition method based on industrial Internet according to claim 3 is characterized in that: The steps to form quality test results include: comparing the quality test distribution map with the quality test threshold; Marking the qualified status of the product to be tested; Calculate the average qualified rate of the products to be tested, and compare the average qualified rate with the qualified rate threshold to form a corresponding quality test result; Wherein, the quality test threshold is related to the preset distance and / or the number of the voice receivers; The average qualified rate is the proportion of qualified products in all the products to be tested; The pass rate threshold is the minimum standard of the pass rate corresponding to the quality test.

5. The artificial intelligence speech recognition method based on industrial Internet according to claim 4 is characterized in that: When the average qualified rate is greater than the qualified rate threshold, the product to be inspected is shipped out of the factory, and when the average qualified rate is less than the qualified rate threshold, the position of the trigger element is adjusted accordingly.

6. The artificial intelligence speech recognition method based on industrial Internet according to claim 5 is characterized in that: The product to be tested is quality tested again, and the corresponding secondary qualified rate is counted. When the secondary qualified rate is greater than the qualified rate threshold, the corresponding test result is marked as passed, and the product to be tested is shipped out of the factory.

7. The artificial intelligence speech recognition method based on industrial Internet according to claim 6 is characterized in that: When the secondary pass rate is less than the pass rate threshold, the corresponding test result is marked as failed, and the voice receiver responds to the trigger instruction to issue a focusing instruction to the recognition device, determines the object distance between the recognition device and the product to be detected, and focuses the recognition device based on the mapping relationship between the preset focus position and the object distance, and takes a high-definition image corresponding to the faulty product in the quality test distribution map, and uploads it to the recognition platform.

Images