An experimental table teaching system based on artificial intelligence
By designing an AI-based experimental teaching system, the problems of large equipment space occupation and noise interference were solved, enabling synchronous teaching between teachers and students, providing personalized feedback and real-time guidance, and improving teaching efficiency and effectiveness.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- SICHUAN DINGZHIHUI TEACHING EQUIP CO LTD
- Filing Date
- 2023-10-08
- Publication Date
- 2026-05-29
AI Technical Summary
Existing AI education teaching and experimental equipment occupies a large space, cannot provide real-time personalized feedback and guidance, and poses safety hazards and noise interference in the experimental environment.
Design an AI-based experimental teaching system, including an operation module, a speech processing module, and a speech control module. The system utilizes preprocessing, feature extraction, and decision models for speech signal processing, and combines HMM algorithm and BP neural network for speech recognition to achieve synchronous teaching between teachers and students. The speech processing module also removes noise interference and generates personalized feedback.
It improved teaching efficiency, ensured the safety of the experimental environment, enabled personalized feedback and real-time guidance, reduced the impact of noise interference on voice control, and enhanced teaching effectiveness.
Smart Images
Figure CN117373296B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of experimental teaching equipment technology, specifically to an artificial intelligence-based experimental teaching system. Background Technology
[0002] Experimental workbenches are platforms used by hospitals, schools, chemical plants, research institutes, and other organizations for conducting experiments, testing, and storing instruments. Artificial intelligence education emphasizes the integration of theory and practice, and experimental workbenches play a crucial supporting role in the practical teaching of artificial intelligence.
[0003] However, in current AI education teaching experimental platforms on the market, the experimental platform, control system, robot, and other teaching aids are set up separately. This setup takes up a lot of space and is limited in actual operation, which is not conducive to AI education.
[0004] Furthermore, due to limitations in equipment and teacher resources, students often cannot receive real-time and personalized feedback and guidance. Specifically, for safety reasons, students cannot perform the experiments during teacher demonstrations and must do so after the demonstration is complete. This lack of targeted feedback and guidance leads to unsatisfactory experimental results. Additionally, the use of chemistry lab benches generates some hazardous substances, requiring strict management of ventilation and waste disposal. Summary of the Invention
[0005] The purpose of this invention is to provide an artificial intelligence-based experimental teaching system for teachers to conduct simultaneous demonstration teaching during language instruction.
[0006] This invention is achieved through the following technical solution:
[0007] An AI-based experimental teaching system includes an experimental platform, an operation module, a speech processing module, and a speech control module. The operation module is mounted on the experimental platform and is used for teaching experiment demonstrations. The speech processing module is used for speech input and processing, generating feature signals after speech processing. The speech control module is signal-connected to the speech processing module and is used to generate command signals based on the feature signals. The operation module is signal-connected to the speech control module and is used to perform teaching experiment demonstrations based on the command signals. The speech processing module includes a preprocessing unit, a feature extraction unit, and a decision model unit connected by signals. The preprocessing unit performs speech preprocessing sequentially including: pre-emphasis, endpoint detection, and frame segmentation with Hamming window processing. The feature extraction unit performs feature extraction sequentially including: calculating MFCC coefficients and calculating the coefficients combining MFCC and LPCC. The decision model unit trains a matching template based on an HMM algorithm model and then obtains the feature signals through the matching template. In the HMM algorithm model, all diagonal elements of the state transition matrix are 0, meaning that each state has no self-transition arcs.
[0008] It should be noted that in existing technologies, due to limitations in equipment and teacher resources, students typically cannot receive real-time and personalized feedback and guidance. Specifically, for safety reasons, students cannot perform the experiments during teacher demonstrations; they must do so after the demonstration. This lack of targeted feedback and guidance leads to unsatisfactory experimental results. Although some mechanical systems with voice control exist, they have not been used in laboratory teaching environments. Furthermore, the complex environment of laboratory teaching, including but not limited to student speech noise, student movement noise, and equipment noise, can all affect voice control. Combined with the real-time nature of experimental operations, this poses a risk of incidental accidents. To address these issues, an AI-based experimental teaching system is proposed. This system separates teacher demonstration experiments from theoretical teaching experiments. The experimental teaching system performs demonstration experiments while the teacher conducts theoretical teaching experiments simultaneously, and students can also follow the system's demonstrations. During this process, the teacher can switch between theoretical teaching and supervision, ensuring a safe experimental environment and significantly improving teaching efficiency. Furthermore, addressing the issues of excessive noise interference and insensitive control in traditional voice control systems applied in experimental teaching environments, a voice processing module is used to process the input voice. The processing includes pre-emphasis, endpoint detection, and frame-by-frame Hamming window processing. During the processing, the diagonal elements of the state transition matrix are improved to 0, meaning that there are no self-transition arcs in each state. This overcomes the limitation of the duration of the traditional HMM algorithm model, thereby highlighting the characteristics of the voice signal.
[0009] Furthermore, the speech processing module also includes a time delay partitioning unit connected to the preprocessing unit. The time delay partitioning unit uses the time point corresponding to the peak of the cross-correlation function as the time delay between the speech input signals to obtain the preprocessing interval. The speech preprocessing process is performed within the preprocessing interval. It should be noted that in experimental teaching environments, there is considerable noise interference, such as student speech noise, student action noise, and equipment operating noise. Traditional noise processing in speech control does not process the actual noise value within the speech segment, but rather the average noise value of the segment before the speech start point. However, since the actual noise is not stationary but randomly changing, the denoised speech signal retains some residual noise between different frames due to the different noise levels in each frame. In contrast to the above methods, this application uses a time delay partitioning unit to partition the time delay between speech signals arriving at the speech processing module, thereby separating interval units that do not contain feature command signals. Interval units without feature command signals are not directly processed; interval units containing feature command signals undergo subsequent processing to separate the feature signals.
[0010] Furthermore, the preprocessing unit performs endpoint detection based on a dual-threshold endpoint detection method within a preprocessing interval. It should be noted that in processes such as automatic speech recognition that require extracting speech features from speech signals, accurate detection of speech endpoints is a crucial factor affecting system performance. Therefore, within a preprocessing interval, the dual-threshold endpoint detection method is used to analyze the short-time zero-crossing rate to determine the speech situation, thereby identifying the endpoints of the feature signal interval.
[0011] Furthermore, the speech processing module also includes a speech denoising unit connected to the preprocessing unit. This unit performs a weighted average of the multi-window power spectral density estimate of the current frame of the speech signal between adjacent frames. The main function of the speech denoising unit is to process the speech signal and reduce noise. Specifically, it estimates the power spectral density of each frame of the speech signal. Power spectral density is a way to quantify the frequency content of a signal; it describes the energy distribution of the signal at different frequencies. This estimate helps us understand the noise components in the speech signal. The weighted average between adjacent frames aims to reduce the impact of random noise on a single frame by averaging information from multiple frames. The weighted average assigns higher weights to more recent frames. In summary, by performing spectral analysis and weighted averaging on the speech signal, the quality and intelligibility of the speech signal are improved.
[0012] Furthermore, the decision model unit also incorporates a BP neural network model, using the cumulative probabilities of all states in the HMM algorithm model as input. The decision-making process of the decision model unit is as follows: the number of hidden states in the HMM algorithm model is set to 4, and the HMM algorithm model is initialized; the HMM algorithm model is trained using training samples to obtain various types of HMM algorithm models, i.e., matching templates; the extracted speech signal is then input into the matching templates, and the cumulative output probabilities under each type of HMM algorithm model are decoded and solved; finally, the cumulative output probabilities are normalized, and the processed data is input into the BP neural network model for recognition decision to obtain the feature signal. It should be noted that the HMM algorithm model has a relatively weak classification ability during speech signal output. By combining it with another neural network model, when there is an error between the actual output signal and the expected output signal in the model, this hybrid model can update the weights and thresholds according to the error, thus improving the performance of the speech recognition system and laying the foundation for accurate control and recognition by the operation module.
[0013] Furthermore, the teaching system also includes a projection module and a teaching module connected by a signal. The teaching module includes a teaching feedback generation unit, used to generate corresponding teaching feedback based on the student's voice or text input. The teaching feedback generation unit further includes a feedback generation algorithm, which is based on deep learning for natural language generation and uses the Transformer model. Specifically, assuming the input sequence is x = [x1, x2, ..., xn], where xi represents the i-th element in the sequence, a word or a character, the algorithm searches for a sequence x that can map to the output sequence y = [y1, y2, ..., ...]. The function f of ym], where yj represents the j-th element in the output sequence, is defined in the Transformer model as: where Qi, Ki, and Vi are the query, key, and value, respectively, obtained by multiplying the learned matrices Wq, Wk, and Wv with the input xi. "." represents matrix multiplication; "T" represents transpose; "sqrt" represents square root; and dk represents the dimension of the key. This formula is used to calculate each element in the output sequence. The projection module is used to project the teaching feedback generated within the teaching module.
[0014] It should be noted that, for the teaching module, the teaching module will collect a series of inputs, including students' questions or instructions, the current state of the experiment, and other relevant information. Then, these inputs will be converted into a sequence x, which will be input into the model. The model will generate an output sequence y, which represents the feedback that the system should give. Finally, this feedback will be converted into human-understandable text and then displayed to the students through the projection module.
[0015] Compared with the prior art, the present invention has the following advantages and beneficial effects:
[0016] 1. This invention separates the teacher's operational demonstration experiments from the theoretical teaching experiments. That is, the experimental teaching system conducts operational demonstration experiments, the teacher conducts theoretical teaching experiments simultaneously, and students can also follow the experimental teaching system to conduct actual demonstrations. In this process, the teacher can switch between theoretical teaching and supervision, ensuring the safety of the experimental environment and greatly improving teaching efficiency.
[0017] 2. This invention also addresses the problem that traditional voice control systems suffer from significant noise interference and insensitive control when applied in experimental teaching environments. It processes the input voice through a voice processing module, including pre-emphasis, endpoint detection, and frame-by-frame Hamming window processing. During the processing, by improving the diagonal elements of the state transition matrix to 0, i.e., each state has no self-transition arc, the problem of the traditional HMM algorithm model having a large limitation in duration is overcome, thereby highlighting the characteristics of the voice signal more.
[0018] 3. This invention, through its teaching module, can understand and process students' questions and needs in real time, generating personalized feedback and guidance, thereby helping students better understand and master experimental knowledge and improving teaching efficiency and effectiveness. Attached Figure Description
[0019] The accompanying drawings, which are included to provide a further understanding of embodiments of the invention and form part of this application, do not constitute a limitation thereof. In the drawings:
[0020] Figure 1 This is a schematic diagram of the system block structure of Embodiment 1 of the present invention;
[0021] Figure 2 This is a schematic diagram of the system block structure of Embodiment 2 of the present invention;
[0022] Figure 3 This is a schematic diagram of the structure of Embodiment 5 of the present invention;
[0023] Figure 4 This is a schematic diagram of the backlash-free worm gear in Embodiment 5 of the present invention.
[0024] The attached diagram shows the markings and corresponding component names:
[0025] 1-Experimental platform, 2-Operation module, 3-Operation unit, 4-Mechanical arm, 5-Clamping component, 6-Bionic component, 41-Worm gear, 42-Fixed part, 43-Moving part, 44-Elastic part, 45-Adjusting part, 441-Spring, 451-Adjusting bolt, 452-Adjusting pad. Detailed Implementation
[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments and accompanying drawings. The illustrative embodiments and descriptions of this invention are for illustrative purposes only and are not intended to limit the invention. It should be noted that this invention is already in the actual research and development stage.
[0027] Example 1:
[0028] As attached Figure 1 As shown, an artificial intelligence-based experimental teaching system includes an experimental platform 1, an operation module 2, a voice processing module, and a voice control module.
[0029] The operation module 2 is set on the experimental table 1 and is used for teaching experiment demonstrations;
[0030] The voice processing module is used for voice input and voice processing, and generates feature signals after voice processing is completed.
[0031] The voice control module is signal-connected to the voice processing module and is used to generate command signals based on the feature signals;
[0032] The operation module 2 is signal-connected to the voice control module and is used to perform teaching experiment demonstrations according to the instruction signals;
[0033] The speech processing module includes a preprocessing unit for signal connection, a feature extraction unit, and a decision model unit. The speech preprocessing process performed by the preprocessing unit includes, in sequence, pre-emphasis, endpoint detection, and frame segmentation with Hamming window processing. The feature extraction process performed by the feature extraction unit includes, in sequence, calculating MFCC coefficients and calculating the coefficients combining MFCC and LPCC. The decision model unit trains a matching template based on the HMM algorithm model and then obtains the feature signal through the matching template. In the HMM algorithm model, all diagonal elements of the state transition matrix are 0, that is, each state has no self-transition arc.
[0034] It should be noted that although some mechanical systems with voice control exist in the existing technology, they have not been used in laboratory teaching environments. Furthermore, the complex environment of laboratory teaching, including but not limited to student speech noise, student movement noise, and equipment operating noise, can all affect the voice control process. Combined with the real-time nature of experimental operations, this poses a risk of derivative accidents. To address these issues, an artificial intelligence-based experimental teaching system is proposed. This system separates teacher demonstration experiments from theoretical teaching experiments. Specifically, the experimental platform 1 teaching system conducts demonstration experiments while the teacher simultaneously conducts theoretical teaching experiments, and students can also simultaneously follow the experimental platform 1 teaching system for actual demonstrations. During this process, the teacher can switch between theoretical teaching and supervision, ensuring the safety of the experimental environment and greatly improving teaching efficiency. Furthermore, addressing the issues of excessive noise interference and insensitive control in traditional voice control systems applied in experimental teaching environments, a voice processing module is used to process the input voice. The processing includes pre-emphasis, endpoint detection, and frame-by-frame Hamming window processing. During the processing, the diagonal elements of the state transition matrix are improved to 0, meaning that there are no self-transition arcs in each state. This overcomes the limitation of the duration of the traditional HMM algorithm model, thereby highlighting the characteristics of the voice signal.
[0035] The voice processing module necessarily includes a voice input unit connected to the preprocessing unit signal, including but not limited to microphone devices. This voice input unit can be fixedly placed on the experimental platform 1, or a mobile voice input unit can be used, or a combination of both. The voice control module is preferably integrated with the multimedia teaching equipment unit, i.e., a PC main control system. During control, it can receive characteristic command signals recognized by the voice processing module, determine the corresponding action to be performed by the operation module 2 based on the received characteristic command signals, and complete the pulse distribution to each degree of freedom servo motor during the determination process.
[0036] It should be noted that the speech processing module also includes a time delay partitioning unit connected to the preprocessing unit. This time delay partitioning unit uses the time point corresponding to the peak of the cross-correlation function as the time delay between the speech input signals to obtain the preprocessing interval. The speech preprocessing process is performed within this preprocessing interval. It should also be noted that in experimental teaching environments, there is considerable noise interference, such as student speech noise, student movement noise, and equipment operating noise. Traditional noise processing in speech control does not process the actual noise value within the speech segment, but rather the average noise value of the segment before the speech start point. However, since the actual noise is not stationary but randomly changing, this results in residual noise remaining between different frames in the denoised speech signal due to the different noise levels in each frame.
[0037] Unlike the methods described above, in this application, the time delay between speech signals arriving at the speech processing module is partitioned by a time delay partitioning unit, thereby separating the interval units that do not contain feature command signals. The interval units that do not contain feature command signals are not directly processed, while the interval units that contain feature command signals are processed to separate the feature signals. For the time-delay partitioning unit, the preferred method for sound source localization is based on the generalized cross-correlation function in the TDOA estimation method. First, the relative time delay from the sound source to the speech input unit is estimated. Then, the distance difference between each sound source is estimated using the time delay value. Finally, a geometric or search algorithm is used to determine whether the corresponding interval unit contains the characteristic command signal. More specifically, first, the speech input unit performs a speech input operation to complete the acquisition of speech information, thereby obtaining the time-domain waveform of the speech signal. Second, the time-domain of the speech signal is transformed to the frequency domain, and the cross-power spectrum between each pair of received signals is obtained by multiplying them pairwise. Then, a weighting operation is performed on the cross-power spectrum to reduce the adverse effects of noise and reverberation in the experimental teaching environment. Finally, the frequency domain of the received signal is inversely transformed to the time domain to obtain the cross-correlation function between each pair of received signals. At this point, the peak value of the cross-correlation function corresponds to the time delay between the received signals. The time-delay partitioning unit performs time-delay partitioning using this time delay, which can greatly eliminate interval units that do not contain the characteristic command signal, unlike the process of processing the entire speech segment in existing technologies.
[0038] It should be noted that the preprocessing unit performs endpoint detection based on a dual-threshold endpoint detection method within a preprocessing interval. It should also be noted that in processes such as automatic speech recognition that require extracting speech features from speech signals, accurate detection of speech endpoints is a crucial factor affecting system performance. Therefore, within a preprocessing interval, the dual-threshold endpoint detection method is used to analyze short-time zero-crossing rates to determine the speech situation and thus identify the endpoints of the feature signal interval. Due to the large number of students, complex noise environments, and significant similarity between noise signals with potentially long durations in experimental teaching environments, traditional dual-threshold endpoint detection methods, which rely on short-time energy to detect such noise, may miss some instances. Based on the above, the preprocessing unit also analyzes short-time zero-crossing rates to determine noise conditions and, in addition, considers short-time average amplitude for endpoint detection. More specifically, within a preprocessing interval, a value is set: the threshold value is the short-time energy value, and the threshold value is the short-time over-threshold rate value. First, in the silent segment of the signal, if the set lower threshold value is lower than the previously set short-time energy threshold value and short-time over-threshold rate value, the starting point is marked. When both parameters return directly below the threshold, the marked starting point is eliminated. If one of the two parameters exceeds the set higher threshold value, it is created as a new starting point, and the detection continues. If both parameters are less than the lower threshold value, the time taken from the starting point to this point is detected. If the detection time is very short, it can be determined as noise, the starting point mark is canceled, and the detection restarts. If the detection time requirement is met, an ending point mark is created.
[0039] Example 2:
[0040] This embodiment only describes the parts that differ from Embodiment 1, specifically:
[0041] As attached Figure 2 As shown, the speech processing module also includes a speech denoising unit connected to the preprocessing unit. This unit performs a weighted average of the multi-window power spectral density estimate of the current frame of the speech signal between adjacent frames. The main function of the speech denoising unit is to process the speech signal and reduce noise. Specifically, it estimates the power spectral density of each frame of the speech signal. Power spectral density is a way to quantify the frequency content of a signal; it describes the energy distribution of the signal at different frequencies. This estimate helps us understand the noise components in the speech signal. The speech denoising unit performs a weighted average between adjacent frames to reduce the impact of random noise on a single frame by averaging information from multiple frames. The weighted average assigns higher weights to more recent frames. In summary, by performing spectral analysis and weighted averaging on the speech signal, the quality and intelligibility of the speech signal are improved.
[0042] It should also be noted that, based on the above, the following steps can reduce impulse-type noise interference: Step 1, input the speech signal through the speech input unit, and then window and frame it, setting the overlap between frames to 50%; Step 2, perform a Fast Fourier Transform on the framed speech signal, and obtain the amplitude spectrum and phase spectrum respectively; Step 3, perform multi-window spectrum estimation on the framed signal, and obtain the multi-window spectrum power spectral density by using a function; Step 4, smooth the multi-window spectrum power spectral density in Step 3; Step 5, obtain the average power spectral density value of the noise segment; Step 6, calculate the gain factor using the spectral subtraction relationship; Step 7, obtain the amplitude spectrum after spectral subtraction using the average amplitude and gain factor. Based on the amplitude spectrum after spectral subtraction and the phase spectrum obtained in Step 2, perform an inverse Fast Fourier Transform to restore the frequency domain to the time domain, thus obtaining the denoised speech signal. Based on the above steps, the denoised speech signal can better restore the waveform of the original speech signal and can improve the signal-to-noise ratio.
[0043] Example 3:
[0044] This embodiment only describes the parts that differ from Embodiment 1, specifically:
[0045] It should be noted that the decision model unit is also equipped with a BP neural network model, and uses the cumulative probability of all states in the HMM algorithm model as the input of the BP neural network model.
[0046] The decision-making process of the decision model unit is as follows: the number of hidden states of the HMM algorithm model is set to 4, and the HMM algorithm model is initialized; the HMM algorithm model is trained with training samples to obtain various types of HMM algorithm models, i.e., matching templates; the speech signal after feature extraction is input into the matching template, and the cumulative output probability under each type of HMM algorithm model is decoded and solved; finally, the cumulative output probability is normalized and the processed data is input into the BP neural network model for recognition decision to obtain feature signals.
[0047] It should also be noted that the HMM algorithm model has a relatively weak classification ability during speech signal output. By combining it with another neural network model, when there is an error between the actual output signal and the expected output signal, this hybrid model can update the weights and thresholds based on the error, thus improving the performance of the speech recognition system and laying the foundation for accurate control and recognition by the operation module 2. The training process includes: setting the number of neurons in each layer of the BP neural network model, the expected error, and the maximum number of network iterations; initializing the network weights and thresholds; calculating the actual network output and the error between the actual output and the expected output; correcting the network weights and thresholds based on this error; and determining whether the error meets the requirements. The training stops when the error meets the requirements or the training iterations reach the set maximum number of network iterations. Based on the above hybrid model, a higher recognition rate can be achieved.
[0048] Example 4:
[0049] This embodiment only describes the parts that differ from Embodiment 1, specifically:
[0050] It should be noted that the teaching system also includes a projection module and a teaching module connected by a signal. The teaching module includes a teaching feedback generation unit, used to generate corresponding teaching feedback based on the student's voice or text input. The teaching feedback generation unit further includes a feedback generation algorithm, which is based on deep learning for natural language generation and uses the Transformer model. Specifically:
[0051] Suppose the input sequence to be processed is x = [x1, x2, ..., xn], where xi represents the i-th element in the sequence, a word or a character. We seek a function f that maps sequence x to the output sequence y = [y1, y2, ..., ym], where yj represents the j-th element in the output sequence. In the Transformer model, this function f is defined as follows:
[0052] Where Qi, Ki, and Vi are the query, key, and value, respectively, obtained by multiplying the learned matrices Wq, Wk, and Wv with the input xi.
[0053] "." indicates matrix multiplication;
[0054] “T” indicates transpose;
[0055] “sqrt” represents the square root;
[0056] dk represents the dimension of the key, and this formula is used to calculate each element in the output sequence;
[0057] The projection module is used to project the teaching feedback generated within the teaching module.
[0058] It's also worth noting that the teaching module collects a series of inputs, including student questions or instructions, the current state of the experiment, and other relevant information. These inputs are then converted into a sequence x, which is fed into the model. The model generates an output sequence y, representing the feedback the system should provide. Finally, this feedback is converted into human-understandable text and displayed to the student via the projection module. For example, if a student asks, "How much hydrochloric acid should I add now?" the system might collect the following inputs: the student's question, the experimental steps, and the student's previous actions. These inputs are converted into a sequence x, which is then fed into the model. The model generates an output sequence y, such as "You should add 5 ml of hydrochloric acid," which is then displayed to the student. The teaching module can understand and process student questions and needs in real time, generating personalized feedback and guidance to help students better understand and master experimental knowledge, thereby improving teaching efficiency and effectiveness.
[0059] Example 5:
[0060] This embodiment only describes the parts that differ from Embodiment 1. For ease of understanding, this embodiment provides an implementation method for the operation module 2, specifically as shown in the attached diagram. Figure 3 , 4 As shown, the operation module 2 includes two operation units 3, each operation unit 3 including a robotic arm 4, the end of which is provided with a clamping part 5 or a bionic part 6. The two operation units 3 are spaced apart and can cooperate to complete the experimental operation demonstration process.
[0061] It should be noted that, due to the requirements of the experimental operating environment, the traditional robotic arm 4 structure exhibits vibration during transmission, which can affect the demonstration process and pose certain safety hazards. Therefore, the operation demonstration of the operating unit 3 needs to be carried out smoothly. Based on this, the applicant has improved the traditional robotic arm 4 structure. The internal structure of the robotic arm 4 uses conventional technologies and will not be described in detail here. Only the improved parts will be described. Specifically, the transmission structure inside the robotic arm 4 adopts a worm gear 41 structure. Unlike existing technologies, the worm gear 41 has a backlash elimination structure. The worm gear 41 includes a fixed part 42, a movable part 43, an elastic part 44, and an adjusting part 45. The adjusting part 45 includes an adjusting bolt 451 and an adjusting shim 452. The movable part 43 is hollow inside. One end of the fixed part 42 has an internal thread, and the other end is connected to the output end of the power mechanism. The power mechanism includes, but is not limited to, a motor. The fixed part 42 and the movable part 43 are axially connected by the adjusting bolt 451. The adjusting shim 452 is disposed between the connection between the fixed part 42 and the movable part 43. The elastic part 44 is disposed on the outer periphery of the adjusting bolt 451. The elastic part 44 is preferably a spring 441. The fixed part 42 and the movable part 43 are relatively fixed in position and can rotate synchronously in the working state. An angle sensor is also provided inside the robotic arm 4. The angle sensor is installed at the end of the fixed part 42. Based on the above structure, a drive control axis system can be formed and set inside the robotic arm 4 for dynamic control. The worm gear 41 meshes with the worm wheel to form a spatial conjugate motion suit, and forms a transmission form driven by the azimuth axis drive motor. Then, based on the angle sensor, a measurement feedback unit is formed, which together form a closed-loop control system. This system can achieve precise and smooth transmission of the robotic arm 4 from both mechanical structure and control system aspects, and its dynamic transmission performance can be greatly improved.
[0062] The clamping component 5 and the bionic component 6 are preferably detachably connected to the end of the robotic arm 4. The mechanical structure is relatively conventional and does not involve any changes to the technical solution of this application, so it will not be described in detail here.
[0063] Example 6:
[0064] This embodiment only describes the parts that differ from Embodiment 5. To further improve the stability of the operating unit 3, the gear structure within the operating unit 3 is specifically improved. Specifically, the gear includes: a main gear, two auxiliary gears, two washers, and two nuts. The two auxiliary gears, washers, and nuts are symmetrically arranged at the left and right ends of the main gear, with the gaps filled with butyl rubber. There is a pre-rotation angle between the two auxiliary gears and the main gear. During assembly, after the two auxiliary gears and the main gear have rotated past the pre-rotation angle, their teeth are aligned before meshing with the next gear. Based on this structure, when the gear enters the working state, the torsional deformation of the gear generates an elastic torsional torque, causing the main gear to be in close contact with one side of the tooth surface of the next gear, while the auxiliary gears at both ends are in close contact with the other side of the tooth surface of the next gear. This eliminates the gap between the tooth sides, ultimately achieving smooth transmission of the robotic arm 4.
[0065] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. An artificial intelligence-based experimental teaching system, comprising an experimental platform (1), characterized in that: It also includes an operation module (2), a voice processing module, and a voice control module. The operation module (2) is set on the experimental platform (1) for teaching experiment demonstration; the operation module is operated by a voice-controlled mechanical system for demonstration experiments; The voice processing module is used for voice input and voice processing, and generates feature signals after voice processing is completed; the voice includes the voice input by teaching staff during theoretical teaching and supervision. The voice control module is signal-connected to the voice processing module and is used to generate command signals based on the feature signals; The operation module (2) is connected to the voice control module and is used to perform teaching experiment demonstrations according to the instruction signals, so that students in the theoretical teaching experiment area can synchronously follow the experimental platform teaching system to perform actual demonstrations. The speech processing module includes a preprocessing unit for signal connection, a feature extraction unit, and a decision model unit. The speech preprocessing process performed by the preprocessing unit includes, in sequence, pre-emphasis, endpoint detection, and frame segmentation with Hamming window processing. The feature extraction process performed by the feature extraction unit includes, in sequence, calculating MFCC coefficients and calculating the coefficients combining MFCC and LPCC. The decision model unit trains a matching template based on the HMM algorithm model and then obtains the feature signal through the matching template. In the HMM algorithm model, all diagonal elements of the state transition matrix are 0, that is, each state has no self-transition arc. The speech processing module further includes a time delay partitioning unit that is signal-connected to the preprocessing unit. The time delay partitioning unit uses the time point corresponding to the peak of the cross-correlation function as the time delay between the speech input signals to obtain the preprocessing interval. The speech preprocessing process is performed within the preprocessing interval. The speech processing module also includes a speech input unit, which acquires speech information and obtains the time-domain waveform of the speech signal. The speech processing module transforms the time-domain speech signal to the frequency domain, multiplies them pairwise to obtain the cross-power spectrum between each pair of received signals, performs a weighting operation on the cross-power spectrum to reduce the adverse effects of noise and reverberation in the experimental teaching environment, and then transforms the frequency domain of the received signal back to the time domain to obtain the cross-correlation function between each pair of received signals. The peak value of the cross-correlation function corresponds to the time delay between the received signals, and the time delay partitioning unit performs time delay partitioning based on this time delay.
2. The experimental teaching system based on artificial intelligence according to claim 1, characterized in that: The preprocessing unit performs endpoint detection based on the dual-threshold endpoint detection method within a preprocessing interval.
3. The experimental teaching system based on artificial intelligence according to claim 1, characterized in that: The speech processing module further includes a speech noise reduction unit that is signal-connected to the preprocessing unit. The speech noise reduction unit is used to perform a weighted average of the multi-window spectral power spectral density estimate of the current frame of the speech signal between adjacent frames.
4. The experimental teaching system based on artificial intelligence according to claim 1, characterized in that: The decision model unit is also equipped with a BP neural network model, and uses the cumulative probability of all states in the HMM algorithm model as the input of the BP neural network model.
5. The experimental teaching system based on artificial intelligence according to claim 4, characterized in that: The decision-making process of the decision model unit is as follows: the number of hidden states of the HMM algorithm model is set to 4, and the HMM algorithm model is initialized; the HMM algorithm model is trained with training samples to obtain various types of HMM algorithm models, i.e., matching templates; the speech signal after feature extraction is input into the matching template, and the cumulative output probability under each type of HMM algorithm model is decoded and solved; finally, the cumulative output probability is normalized and the processed data is input into the BP neural network model for recognition decision to obtain feature signals.