Electric toothbrush system based on effect analysis of artificial intelligence

Abstract

The invention relates to the technical field of intelligent electric toothbrushes, in particular to an electric toothbrush system based on effect analysis of artificial intelligence, which comprises a multi-source data acquisition module, an AI algorithm processing module, a dynamic effect optimization module and a user interaction feedback module, and is characterized in that the multi-source data acquisition module acquires operation data of an electric toothbrush; the AI algorithm processing module receives the operation data, performs intelligent analysis and outputs a tooth brushing effect analysis result; the dynamic effect optimization module receives the tooth brushing effect analysis result and adjusts operation parameters of the toothbrush; the user interaction feedback module receives the tooth brushing effect analysis result and the operation parameter information and outputs the result to a user; and the AI algorithm processing module adopts a federal transfer learning architecture, updates an analysis model based on local data, and performs encryption collaborative iteration with the cloud server. According to the invention, through federal transfer learning, vibration spectrum identification and no-load calibration and mixed model double feedback, model optimization under privacy protection, accurate identification and error correction and multi-dimensional improvement of tooth brushing effect are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent electric toothbrush technology, and in particular to an electric toothbrush system based on artificial intelligence effect analysis. Background Technology

[0002] The technological development of electric toothbrushes has consistently focused on improving cleaning performance and user experience. Early products were primarily mechanically driven, mainly meeting basic cleaning needs by optimizing vibration frequency and the variety of cleaning modes. With the increasing maturity of sensing technology, components such as pressure sensors and accelerometers have been integrated into products, enabling electric toothbrushes to collect basic operational data during brushing and providing simple feedback functions such as pressure reminders and brushing time tracking. In recent years, the continued penetration of artificial intelligence and the Internet of Things (IoT) technologies has become an important direction for industry development. Some products have begun to incorporate machine learning models for preliminary evaluation of brushing effectiveness, while simultaneously establishing data interaction channels with cloud platforms to expand into new functions such as intelligent monitoring and usage data recording, driving the transformation of electric toothbrushes from simple cleaning tools into intelligent health monitoring products.

[0003] The current application of smart electric toothbrush technology still has many significant shortcomings. Regarding data security, some products require uploading users' raw brushing data to the cloud for functional upgrades or model optimization, a process that poses a significant risk of data leakage. The brush head adaptation process lacks efficient and accurate automatic recognition methods, mostly relying on users manually selecting the model. This not only increases operational complexity but may also affect cleaning effectiveness and data analysis accuracy due to incorrect model matching. Sensors are susceptible to performance drift due to factors such as brush head wear and environmental humidity during long-term use, and existing products generally lack targeted real-time calibration measures, leading to deviations in data collection and analysis results. Furthermore, existing analysis models are mostly limited to single-dimensional evaluation, failing to comprehensively cover key indicators such as brushing force, cleaning area coverage, and brushing posture. Model updates also lack flexible adaptation mechanisms, failing to fully adapt to different users' habits, and feedback functions often suffer from insufficient real-time performance or unintuitive information presentation, making it difficult to effectively help users improve their brushing quality. Summary of the Invention

[0004] The purpose of this invention is to overcome the above-mentioned problems and provide an electric toothbrush system based on artificial intelligence effect analysis. To achieve the above objective, this invention adopts the following technical solution:

[0005] An electric toothbrush system based on artificial intelligence-driven effect analysis includes a multi-source data acquisition module, an AI algorithm processing module, a dynamic effect optimization module, and a user interaction feedback module. The multi-source data acquisition module collects operational data from the electric toothbrush. The AI ​​algorithm processing module communicates with the multi-source data acquisition module, receives the operational data, performs intelligent analysis, and outputs brushing effect analysis results. The dynamic effect optimization module communicates with the AI ​​algorithm processing module, receives the brushing effect analysis results, and adjusts the electric toothbrush's operating parameters based on the results. The user interaction feedback module communicates with both the AI ​​algorithm processing module and the dynamic effect optimization module, receives the brushing effect analysis results and the adjusted operating parameters, and outputs them to the user. The AI ​​algorithm processing module employs a federated transfer learning architecture, updating its analysis model based on local data and performing encrypted collaborative iteration with a cloud server.

[0006] Furthermore, the AI ​​algorithm processing module includes a vibration spectrum recognition module, a real-time effect analysis module, and a federated transfer learning module; the vibration spectrum recognition module performs spectrum analysis on the vibration signals in the running data to identify the current brush head type; the real-time effect analysis module performs real-time analysis based on the running data and the identified brush head type through a locally deployed machine learning model to obtain the brushing effect analysis results; the federated transfer learning module performs model updates and collaborative iterations under the federated transfer learning architecture.

[0007] Furthermore, the vibration spectrum identification module performs a fast Fourier transform on the vibration signal, converting it from the time domain to the frequency domain to obtain the vibration spectrum characteristics. The calculation formula is as follows:

[0008] F = FFT(s(t));

[0009] Where F represents the vibration spectrum characteristics, FFT is the fast Fourier transform function, and s(t) is the vibration signal in the time domain;

[0010] The energy distribution of characteristic frequency bands related to the resonance of the physical structure of the brush head is extracted from the vibration spectrum characteristics to form a feature vector; the cosine similarity between the feature vector and the pre-stored reference feature vectors corresponding to different brush head models is calculated, and the brush head type is determined based on the highest similarity result.

[0011] Furthermore, the real-time performance analysis module is deployed on a local microcontroller, employing a hybrid model consisting of a convolutional neural network layer and a converter encoder layer connected in series. The convolutional neural network layer includes convolutional and pooling layers, with input being multidimensional operational data containing temporal information. This multidimensional operational data at least covers vibration frequency, acceleration, and pressure parameters, and is input after standardization preprocessing. Local spatiotemporal features of the data are extracted through convolution and pooling operations, with a linear rectified function used as the activation function. The converter encoder layer includes an attention head and a position encoding unit, performing global attention calculations on the local spatiotemporal features, modeling the dependencies between features at different time points, and setting a dropout mechanism to suppress overfitting. The output layer of the hybrid model outputs at least one evaluation result from brushing force rationality score, brushing area coverage integrity score, and brushing posture correctness judgment result through an activation function.

[0012] Furthermore, the federated transfer learning module locally utilizes the current user's runtime data and corresponding effect analysis result labels to calculate the gradient of the loss function of the machine learning model in the real-time effect analysis module. The calculation formula is as follows:

[0013]

[0014] Where N is the number of data samples used for local training, x i Let y be the input data for the i-th sample. i Let θ be the label data for the i-th sample, θ be the model parameters, and L be the loss function;

[0015] Gradients are encrypted using an asymmetric encryption algorithm. Encryption is performed using a dual mechanism: public key encryption of gradient data combined with private key signature verification. The public key is pre-distributed to the local device from the cloud server, while the private key is stored in the secure storage area of ​​the electric toothbrush and is not transmitted externally. The encrypted gradient and corresponding signature information are uploaded to the cloud server. The updated global model parameters, which are aggregating gradients from multiple devices by the cloud server, are received and the local machine learning model is updated.

[0016] Furthermore, the federated transfer learning module employs a hierarchical parameter tuning strategy, where the parameters θ of the machine learning model are divided into general feature layer parameters θ. g With user adaptation layer parameter θ p During local training and gradient calculation, for θ p Use greater than θ g The learning rate; when aggregating in the cloud, for θ g The gradients are aggregated using a weighted average method, with the weights determined based on the number of local data samples on each device. pThe gradient calculation calculates the variance of gradients across all devices, and based on the variance results, it selects whether to aggregate or retain local updates.

[0017] Furthermore, in the initial stage after each start-up of the electric toothbrush, the dynamic effect optimization module controls the motor to run under no-load and collects the vibration signal at this time as the no-load reference signal. The module performs spectral analysis on the no-load reference signal, selects the resonance peak with the highest energy proportion as the main resonance peak, compares the no-load reference signal with the main resonance peak corresponding to the pre-stored standard no-load signal, and calculates the frequency offset and energy difference between the two. Based on the comprehensive judgment result of the frequency offset and energy difference, the module compensates and adjusts the relevant sensor calibration parameters or model input preprocessing parameters in the AI ​​algorithm processing module.

[0018] Furthermore, the dynamic effect optimization module continuously records the user's brushing effect analysis results and the corresponding operating parameter adjustment history. The brushing effect analysis results include brushing force data, coverage integrity data, and posture error data. The operating parameter adjustment history includes vibration frequency adjustment information and amplitude adjustment information. Based on the historical data within a preset period, a brushing habit profile of the corresponding user is generated through a clustering algorithm. The profile classification is determined based on brushing force characteristics, coverage characteristics, and posture characteristics.

[0019] Furthermore, in the user interaction feedback module, when the brushing effect analysis results indicate that there is excessive brushing force, missed areas, or incorrect posture, the motor of the electric toothbrush is controlled to generate a vibration waveform with a sudden drop in intensity or a regular intermittent vibration waveform to provide real-time tactile alerts. After brushing, a visual report is generated, which includes brushing time, oral cavity zone coverage heatmap, comprehensive effect score, and a comparison of recent brushing effect trends, and is displayed to the user through the bound mobile terminal application.

[0020] The advantages of this invention are:

[0021] 1. This invention employs a federated transfer learning architecture in the AI ​​algorithm processing module, combined with a dual verification mechanism using asymmetric encryption. It divides the model parameters into a general feature layer and a user adaptation layer and adjusts the learning rate differently. Only the model gradient is encrypted and uploaded while the original data is stored locally. This allows the model to improve its analysis accuracy by relying on local data updates and global parameter optimization without compromising user privacy. The model can not only fit individual usage habits but also continuously improve its analytical capabilities.

[0022] 2. This invention uses a vibration spectrum recognition module to perform spectrum analysis on vibration signals to automatically match the brush head type. At the same time, it collects an idle reference signal during the initial stage of electric toothbrush startup, compares it with the standard signal to calculate the frequency offset and energy difference, and compensates for relevant parameters. This enables accurate identification of the brush head model without manual operation by the user, effectively corrects errors caused by sensor drift and brush head wear, and improves the accuracy and stability of brushing effect analysis.

[0023] 3. This invention deploys a hybrid model consisting of a convolutional neural network layer and a converter encoder layer on a local microcontroller to analyze multidimensional data including parameters such as vibration frequency, acceleration, and pressure. At the same time, it sets up a real-time tactile alert and visual report feedback mechanism to achieve multidimensional real-time evaluation of brushing force, coverage area, and brushing posture. This allows users to adjust their behavior in a timely manner during brushing and clearly understand the details and changes in the effect after brushing, thus helping to improve the quality of brushing. Attached Figure Description

[0024] The accompanying drawings, which form part of this application, are used to provide a further understanding of the application and to make other features, objects, and advantages of the application more apparent. The illustrative embodiments and descriptions of this application are used to explain the application and do not constitute an undue limitation of the application.

[0025] In the attached diagram:

[0026] Figure 1 This is the overall architecture diagram of the electric toothbrush system based on artificial intelligence effect analysis in Example 1.

[0027] Figure 2 This is a diagram showing the internal structure of the AI ​​algorithm processing module of the electric toothbrush system based on artificial intelligence effect analysis in Example 1.

[0028] Figure 3 This is a flowchart of the vibration spectrum recognition process for the electric toothbrush system based on artificial intelligence effect analysis in Example 1.

[0029] Figure 4 This is a structural diagram of the real-time effect analysis hybrid model of the electric toothbrush system based on artificial intelligence effect analysis in Example 1.

[0030] Figure 5 This is a flowchart of the dynamic effect optimization and calibration process of the electric toothbrush system based on artificial intelligence effect analysis in Example 1.

[0031] Figure 6 This is a flowchart of the user profile generation process for the electric toothbrush system based on artificial intelligence effect analysis in Example 1.

[0032] Figure 7This is a flowchart of the user interaction feedback process for the electric toothbrush system based on artificial intelligence effect analysis in Example 1. Detailed Implementation

[0033] The present invention will now be described in detail and specifically through specific embodiments in order to better understand the present invention. However, the following embodiments do not limit the scope of protection of the present invention.

[0034] Example 1

[0035] like Figure 1 As shown, the electric toothbrush system based on artificial intelligence for effect analysis includes a multi-source data acquisition module, an AI algorithm processing module, a dynamic effect optimization module, and a user interaction feedback module. The multi-source data acquisition module collects the operating data of the electric toothbrush. The AI ​​algorithm processing module is communicatively connected to the multi-source data acquisition module, receives the operating data, performs intelligent analysis, and outputs the brushing effect analysis results. The dynamic effect optimization module is communicatively connected to the AI ​​algorithm processing module, receives the brushing effect analysis results, and adjusts the operating parameters of the electric toothbrush according to the results. The user interaction feedback module is communicatively connected to the AI ​​algorithm processing module and the dynamic effect optimization module, receives the brushing effect analysis results and the adjusted operating parameter information, and outputs them to the user. The AI ​​algorithm processing module adopts a federated transfer learning architecture, updates its analysis model based on local data, and performs encrypted collaborative iteration with the cloud server.

[0036] In a specific embodiment, each module of the system communicates with the control circuit via an onboard bus. The multi-source data acquisition module uses a six-axis motion tracking sensor and a flexible thin-film pressure sensor to collect operational data such as vibration, pressure, and acceleration, with a sampling frequency of 100Hz. The AI ​​algorithm processing module is equipped with a high-performance 32-bit microcontroller with a main frequency of up to 480MHz and sufficient Flash and RAM to support algorithm operation. The dynamic effect optimization module controls the electric toothbrush motor through a motor drive circuit, and the user interaction feedback module integrates a Bluetooth 5.0 communication unit and a motor control unit. After the system starts, the multi-source data acquisition module transmits the real-time collected operational data to the AI ​​algorithm processing module, which analyzes the data and outputs the brushing effect results. The dynamic effect optimization module adjusts parameters such as motor vibration frequency and amplitude accordingly. The user interaction feedback module synchronizes relevant information to the mobile terminal application via Bluetooth. Simultaneously, the AI ​​algorithm processing module updates the model based on local data and achieves collaborative iteration with the cloud server through an encrypted channel.

[0037] Furthermore, such as Figure 2As shown, the AI ​​algorithm processing module includes a vibration spectrum recognition module, a real-time effect analysis module, and a federated transfer learning module. The vibration spectrum recognition module performs spectrum analysis on the vibration signals in the running data to identify the current brush head type. The real-time effect analysis module performs real-time analysis based on the running data and the identified brush head type using a locally deployed machine learning model to obtain brushing effect analysis results. The federated transfer learning module performs model updates and collaborative iterations under the federated transfer learning architecture.

[0038] In a specific embodiment, the vibration spectrum recognition module, real-time effect analysis module, and federated transfer learning module are all integrated into the microcontroller's software system, and the modules interact with each other through shared memory. The vibration spectrum recognition module prioritizes receiving vibration signals from the multi-source data acquisition module, completes brush head type recognition, and writes the recognition results into shared memory. The real-time effect analysis module reads the running data and brush head type information from shared memory, calls the locally deployed hybrid machine learning model for analysis, and outputs the brushing effect analysis results. The federated transfer learning module periodically reads the model parameters and local training data from the real-time effect analysis module, performs model gradient calculation, encrypted upload, and global parameter update operations, and synchronizes the updated model parameters to the real-time effect analysis module to ensure continuous optimization of the analysis model.

[0039] Furthermore, such as Figure 3 As shown, the vibration spectrum identification module performs a fast Fourier transform on the vibration signal, converting it from the time domain to the frequency domain to obtain the vibration spectrum characteristics. The calculation formula is as follows:

[0040] F = FFT(s(t));

[0041] Where F represents the vibration spectrum characteristics, FFT is the fast Fourier transform function, and s(t) is the vibration signal in the time domain;

[0042] The energy distribution of characteristic frequency bands related to the resonance of the physical structure of the brush head is extracted from the vibration spectrum characteristics to form a feature vector; the cosine similarity between the feature vector and the pre-stored reference feature vectors corresponding to different brush head models is calculated, and the brush head type is determined based on the highest similarity result.

[0043] In a specific embodiment, the vibration spectrum identification module first performs DC component removal and smoothing filtering preprocessing on the received s(t), then calls the microcontroller's built-in FFT operation unit to perform a fast Fourier transform to obtain the vibration spectrum feature F in the frequency domain. Feature frequency bands related to the resonance of the brush head's physical structure within the 100-1000Hz frequency band are extracted, and the energy distribution within these frequency bands is calculated to construct a 128-dimensional feature vector. The microcontroller's Flash memory pre-stores the baseline feature vectors for five common brush head models, with each model corresponding to the mean of 30 sets of calibration data under different environments. Cosine similarity is calculated according to the formula... Execution, in which To collect and construct feature vectors, To store the baseline feature vector, select the model corresponding to the highest similarity value greater than 0.85 as the current brush head type.

[0044] Furthermore, such as Figure 4 As shown, the real-time effect analysis module is deployed on a local microcontroller. The machine learning model used is a hybrid model consisting of a convolutional neural network layer and a converter encoder layer connected in series. The convolutional neural network layer includes convolutional layers and pooling layers. The input is multi-dimensional running data containing temporal information. The multi-dimensional running data at least covers vibration frequency, acceleration, and pressure parameters. After standardization preprocessing, the data is input, and local spatiotemporal features are extracted through convolution and pooling operations. The activation function is a linear rectified function. The converter encoder layer includes an attention head and a position encoding unit. It performs global attention calculation on local spatiotemporal features, models the dependencies between features at different time points, and sets a dropout mechanism to suppress overfitting. The output layer of the hybrid model outputs at least one evaluation result from brushing force rationality score, brushing area coverage integrity score, and brushing posture correctness judgment result through the activation function.

[0045] In a specific embodiment, the real-time performance analysis module is deployed on a high-performance 32-bit microcontroller. The convolutional neural network layer in the hybrid model contains three convolutional layers and two max-pooling layers. The first convolutional layer has a kernel size of 3×3, a number of 32, and a stride of 1; the second convolutional layer has a kernel size of 3×3, a number of 64, and a stride of 1; and the third convolutional layer has a kernel size of 2×2, a number of 128, and a stride of 1. The activation function for all layers is the rectified linear function ReLU(x) = max(0,x). The pooling layers use max pooling with a kernel size of 2×2 and a stride of 2.

[0046] The transformer encoder layer contains two attention heads, has a hidden layer dimension of 256, and uses sinusoidal position encoding. The dropout probability is set to 0.1. Multidimensional operational data must at least cover vibration frequency, acceleration, and pressure parameters, and be preprocessed using standardization. After execution, the input model is set to μ = 0 and σ = 1. The convolutional neural network layer extracts local spatiotemporal features, the transformer-encoder layer models global feature dependencies, and the output layer uses the sigmoid activation function. Output at least one of the following: brushing force rationality score, brushing area coverage integrity score, and brushing posture correctness judgment result. The brushing force rationality score and brushing area coverage integrity score are both in the range of 0-1. In the brushing posture correctness judgment result, 0 represents error and 1 represents correctness.

[0047] Furthermore, the federated transfer learning module locally utilizes the current user's runtime data and corresponding effect analysis result labels to calculate the gradient of the loss function of the machine learning model in the real-time effect analysis module. The calculation formula is as follows:

[0048]

[0049] Where N is the number of data samples used for local training, x i Let y be the input data for the i-th sample. i Let θ be the label data for the i-th sample, θ be the model parameters, and L be the loss function;

[0050] Gradients are encrypted using an asymmetric encryption algorithm. Encryption is performed using a dual mechanism: public key encryption of gradient data combined with private key signature verification. The public key is pre-distributed to the local device from the cloud server, while the private key is stored in the secure storage area of ​​the electric toothbrush and is not transmitted externally. The encrypted gradient and corresponding signature information are uploaded to the cloud server. The updated global model parameters, which are aggregating gradients from multiple devices by the cloud server, are received and the local machine learning model is updated.

[0051] In a specific embodiment, the federated transfer learning module selects N=1000 as the number of data samples used for local training, x i Let y be the input data for the i-th sample, i.e., the multidimensional running data. i Let θ represent the label data of the i-th sample, which corresponds to the brushing effect analysis result, θ represent all parameters of the mixture model, and L uses the cross-entropy loss function. According to the formula Calculate the gradient of the model loss function The RSA-2048 asymmetric encryption algorithm is used to... Encryption is used. The public key is pre-distributed from the cloud server to the secure storage area of ​​the microcontroller via Bluetooth 5.0, while the private key is stored in the trusted platform module of the electric toothbrush and is not transmitted externally. The encrypted gradient and private key signature information are uploaded to the cloud server via the wireless communication module. The local hybrid model is updated by replacing parameters after receiving the global model parameters aggregated from the cloud.

[0052] Furthermore, the federated transfer learning module employs a hierarchical parameter tuning strategy, where the parameters θ of the machine learning model are divided into general feature layer parameters θ. g With user adaptation layer parameter θ p During local training and gradient calculation, for θ p Use greater than θ g The learning rate; when aggregating in the cloud, for θ gThe gradients are aggregated using a weighted average method, with the weights determined based on the number of local data samples on each device. p The gradient calculation calculates the variance of gradients across all devices, and based on the variance results, it selects whether to aggregate or retain local updates.

[0053] In a specific embodiment, the federated transfer learning module employs a hierarchical parameter adjustment strategy, dividing the parameters θ of the machine learning model into general feature layer parameters θ. g With user adaptation layer parameter θ p , where θ g θ represents the common parameters of the first two layers of the convolutional neural network and the transformer encoder layer. p These are the parameters of the third and output layers of the convolutional neural network. During local training, θ g The learning rate is set to 0.002, and θ p The learning rate is set to 0.008 to ensure that θ p Learning rate greater than θ g .

[0054] During the cloud aggregation phase, for θ g gradient according to Execute the aggregation, where M is the number of devices participating in the aggregation, and N... k Let k be the number of local samples for the k-th device. Let θ be the value of the k-th device. g The gradient and weights are determined based on the proportion of local data samples on each device. For θ p Gradient calculation of variance in Let θ be the value of the k-th device. p gradient, For θ p Gradient mean, if σ 2 If σ < 0.05, then average aggregation is performed; if σ 2 If ≥0.05, the local θ of each device will be retained. p Update results.

[0055] Furthermore, such as Figure 5 As shown, in the initial stage after each start-up of the electric toothbrush, the dynamic effect optimization module controls the motor to run under no-load and collects the vibration signal at this time as the no-load reference signal. The no-load reference signal is subjected to spectrum analysis, and the resonance peak with the highest energy ratio is selected as the main resonance peak. The no-load reference signal is compared with the main resonance peak corresponding to the pre-stored standard no-load signal, and the frequency offset and energy difference between the two are calculated. Based on the comprehensive judgment result of frequency offset and energy difference, the relevant sensor calibration parameters or model input preprocessing parameters in the AI ​​algorithm processing module are compensated and adjusted.

[0056] In a specific embodiment, the dynamic effect optimization module controls the motor to run under no-load conditions during the initial stage after each start-up of the electric toothbrush, collecting the vibration signal at this time as the no-load reference signal. Spectral analysis is performed on the no-load reference signal, and a peak detection algorithm is used to select the resonant peaks with the highest energy proportion as the main resonant peaks. A standard no-load signal pre-stored in the microcontroller is then called; this signal is reference data calibrated under standard conditions at the factory. The frequency offset Δf = |f| between the main resonant peaks corresponding to the no-load reference signal and the standard no-load signal is calculated. ref -f meas |E and the energy difference ΔE=|E ref -E meas |, where f ref f is the standard resonant peak frequency. meas E is the resonant peak frequency of the no-load reference signal. ref E is the standard resonance peak energy. meas This represents the resonant peak energy of the unloaded reference signal. Based on the comprehensive judgment result of the frequency offset and energy difference, the relevant sensor calibration parameters or model input preprocessing parameters in the AI ​​algorithm processing module are compensated and adjusted. If Δf > 3Hz or ΔE > E... ref ×15%, according to Compensation sensor calibration parameter K, if Δf≤3Hz and ΔE≤E ref If the value is multiplied by 15%, then the original parameters remain unchanged.

[0057] Furthermore, such as Figure 6 As shown, the dynamic effect optimization module continuously records the user's brushing effect analysis results and the corresponding operating parameter adjustment history. The brushing effect analysis results include brushing force data, coverage integrity data, and posture error data. The operating parameter adjustment history includes vibration frequency adjustment information and amplitude adjustment information. Based on the historical data within a preset period, a brushing habit profile of the corresponding user is generated through a clustering algorithm. The profile classification is determined based on brushing force characteristics, coverage characteristics, and posture characteristics.

[0058] In a specific embodiment, the dynamic effect optimization module continuously records the user's brushing effect analysis results and corresponding operating parameter adjustment history through the onboard EEPROM memory. The brushing effect analysis results include brushing force data, coverage integrity data, and posture error data. The brushing force data includes the average and maximum brushing force, the coverage integrity data includes the coverage time percentage of each area, and the posture error data includes the number of errors and the duration of errors. The operating parameter adjustment history includes vibration frequency adjustment information and amplitude adjustment information, specifically the vibration frequency values ​​and amplitude levels before and after adjustment. Based on historical data within a preset period, a brushing habit profile of the corresponding user is generated through a clustering algorithm. The preset period is set to 30 brushing behaviors, using the K-means clustering algorithm with 3 clusters. The clustering feature dimensions include average brushing force, coverage integrity, posture error frequency, vibration frequency preference, and amplitude preference. The brushing profile classification is determined based on brushing force characteristics, coverage characteristics, and posture characteristics. Specifically, the rules are as follows: A gentle brushing profile is characterized by an average brushing force less than 0.3 times the maximum safe brushing force, coverage completion greater than or equal to 80%, and a posture error frequency of less than 2 times per brushing session; a high-efficiency brushing profile is characterized by an average brushing force between 0.3 and 0.7 times the maximum safe brushing force, coverage completion greater than or equal to 90%, and a posture error frequency of less than 1 time per brushing session; and a brushing profile is characterized by an average brushing force greater than 0.7 times the maximum safe brushing force, or coverage completion less than 80%, or a posture error frequency of greater than or equal to 2 times per brushing session, requiring optimization. The maximum safe brushing force is represented by F. max express.

[0059] Furthermore, such as Figure 7 As shown, in the user interaction feedback module, when the brushing effect analysis results indicate that there is excessive brushing force, missed areas, or incorrect posture, the motor of the electric toothbrush is controlled to generate a vibration waveform with a sudden drop in intensity or a regular intermittent vibration waveform to provide real-time tactile warnings. After brushing, a visual report is generated, which includes brushing time, oral cavity zone coverage heatmap, comprehensive effect score, and a comparison of recent brushing effect trends, and is displayed to the user through the bound mobile terminal application.

[0060] In a specific embodiment, the motor control unit of the user interaction feedback module is connected to the electric toothbrush drive motor, and the Bluetooth 5.0 communication module establishes a stable connection with the user's mobile phone application. When the brushing effect analysis result output by the AI ​​algorithm processing module indicates that the brushing force is too large (i.e., the brushing force value is greater than 0.8 times the maximum safe brushing force), the coverage time of a missed area (i.e., a certain oral cavity area) is less than 5 seconds, or the posture is incorrect, the motor control unit controls the drive motor to generate a vibration waveform with a sudden drop in intensity or a regular intermittent vibration waveform to provide real-time tactile warnings. The parameters of the vibration waveform with a sudden drop in intensity are that the vibration intensity drops from the current value to 30%, lasts for 0.5 seconds, and then recovers to the current value; the parameters of the regular intermittent vibration waveform are that it vibrates for 0.8 seconds, pauses for 0.2 seconds, and cycles twice with a period of 1 second. After brushing, the user interaction feedback module receives the complete analysis results from the AI ​​algorithm processing module and the historical comparison data from the dynamic effect optimization module to generate a visualization report. The report includes brushing time, oral cavity area coverage heatmap, comprehensive effect score, and a comparison of recent brushing effect trends. The oral cavity is divided into maxillary anterior teeth, maxillary posterior teeth, mandibular anterior teeth, mandibular posterior teeth, left buccal side, and right buccal side. The coverage heatmap uses different colors to represent the percentage of coverage time for each area: green indicates a percentage greater than or equal to 20%, yellow indicates a percentage between 10% and 20%, and red indicates a percentage less than 10%. The overall effectiveness score is calculated according to... The calculation is performed, where S is the overall performance score, ranging from 0 to 100, C is the coverage completeness, and F is the overall performance score. avg F represents the average brushing force. max The maximum safe brushing force is represented by E, which indicates the number of incorrect brushing techniques. The recent brushing effect trend comparison is a curve showing the change in the overall effect score over the last 7 brushing sessions. The visual report is transmitted to the user's mobile phone application via Bluetooth 5.0 and displayed to the user.

[0061] The specific embodiments of the present invention have been described in detail above, but they are merely examples, and the present invention is not equivalent to the specific embodiments described above. For those skilled in the art, any equivalent modifications and substitutions to the present invention are also within the scope of the present invention. Therefore, all equivalent transformations and modifications made without departing from the spirit and scope of the present invention should be covered within the scope of the present invention.

Claims

1. An electric toothbrush system based on artificial intelligence-based effect analysis, characterized in that, The device includes a multi-source data acquisition module, an AI algorithm processing module, a dynamic effect optimization module, and a user interaction feedback module. The multi-source data acquisition module collects operational data from the electric toothbrush. The AI ​​algorithm processing module communicates with the multi-source data acquisition module, receives the operational data, performs intelligent analysis, and outputs brushing effect analysis results. The dynamic effect optimization module communicates with the AI ​​algorithm processing module, receives the brushing effect analysis results, and adjusts the electric toothbrush's operational parameters based on the results. The user interaction feedback module communicates with both the AI ​​algorithm processing module and the dynamic effect optimization module, receives the brushing effect analysis results and the adjusted operational parameters, and outputs them to the user. The AI ​​algorithm processing module employs a federated transfer learning architecture, updates its analysis model based on local data, and performs encrypted collaborative iteration with a cloud server.

2. The electric toothbrush system based on artificial intelligence effect analysis according to claim 1, characterized in that, The AI ​​algorithm processing module includes a vibration spectrum recognition module, a real-time effect analysis module, and a federated transfer learning module; the vibration spectrum recognition module performs spectrum analysis on the vibration signals in the running data to identify the current brush head type. The real-time effect analysis module performs real-time analysis based on the running data and the identified brush head type, using a locally deployed machine learning model to obtain brushing effect analysis results. The federated transfer learning module performs model updates and collaborative iterations under the federated transfer learning architecture.

3. The electric toothbrush system based on artificial intelligence effect analysis according to claim 2, characterized in that, The vibration spectrum identification module performs a fast Fourier transform on the vibration signal, converting it from the time domain to the frequency domain to obtain the vibration spectrum characteristics. The calculation formula is as follows: F = FFT(x(t)); Where F represents the vibration spectrum characteristics, FFT is the fast Fourier transform function, and s(t) is the vibration signal in the time domain; The energy distribution of characteristic frequency bands related to the resonance of the physical structure of the brush head is extracted from the vibration spectrum characteristics to form a feature vector; the cosine similarity between the feature vector and the pre-stored reference feature vectors corresponding to different brush head models is calculated, and the brush head type is determined based on the highest similarity result.

4. The electric toothbrush system based on artificial intelligence effect analysis according to claim 3, characterized in that, The real-time performance analysis module is deployed on a local microcontroller and employs a hybrid model consisting of a convolutional neural network layer and a converter encoder layer connected in series. The convolutional neural network layer includes convolutional and pooling layers, and the input is multi-dimensional operational data containing temporal information. This multi-dimensional operational data at least covers vibration frequency, acceleration, and pressure parameters. After standardization preprocessing, the data is input, and local spatiotemporal features are extracted through convolution and pooling operations. The activation function is a linear rectified function. The converter encoder layer includes an attention head and a position encoding unit, which performs global attention calculation on the local spatiotemporal features, models the dependencies between features at different time points, and sets a dropout mechanism to suppress overfitting. The output layer of the hybrid model outputs at least one evaluation result from brushing force rationality score, brushing area coverage integrity score, and brushing posture correctness judgment result through the activation function.

5. The electric toothbrush system based on artificial intelligence effect analysis according to claim 4, characterized in that, The federated transfer learning module uses the current user's runtime data and corresponding effect analysis result labels locally to calculate the gradient of the loss function of the machine learning model in the real-time effect analysis module. The calculation formula is as follows: Where N is the number of data samples used for local training, x i Let y be the input data for the i-th sample. i Let θ be the label data for the i-th sample, θ be the model parameters, and L be the loss function; Gradients are encrypted using an asymmetric encryption algorithm. Encryption is performed using a dual mechanism: public key encryption of gradient data combined with private key signature verification. The public key is pre-distributed to the local device from the cloud server, while the private key is stored in the secure storage area of ​​the electric toothbrush and is not transmitted externally. The encrypted gradient and corresponding signature information are uploaded to the cloud server. The updated global model parameters, which are aggregating gradients from multiple devices by the cloud server, are received and the local machine learning model is updated.

6. The electric toothbrush system based on artificial intelligence effect analysis according to claim 5, characterized in that, The federated transfer learning module employs a hierarchical parameter tuning strategy, and the parameters θ of the machine learning model are divided into general feature layer parameters θ. g With user adaptation layer parameter θ p During local training and gradient calculation, for θ p Use greater than θ g The learning rate; when aggregating in the cloud, for θ g The gradients are aggregated using a weighted average method, with the weights determined based on the number of local data samples on each device. p The gradient calculation calculates the variance of gradients across all devices, and based on the variance results, it selects whether to aggregate or retain local updates.

7. The electric toothbrush system based on artificial intelligence effect analysis according to claim 6, characterized in that, In the initial stage after each start-up of the electric toothbrush, the dynamic effect optimization module controls the motor to run under no-load and collects the vibration signal at this time as the no-load reference signal. The module performs spectral analysis on the no-load reference signal, selects the resonance peak with the highest energy proportion as the main resonance peak, compares the no-load reference signal with the main resonance peak corresponding to the pre-stored standard no-load signal, and calculates the frequency offset and energy difference between the two. Based on the comprehensive judgment result of the frequency offset and energy difference, the module compensates and adjusts the relevant sensor calibration parameters or model input preprocessing parameters in the AI ​​algorithm processing module.

8. The electric toothbrush system based on artificial intelligence effect analysis according to claim 7, characterized in that, The dynamic effect optimization module continuously records the user's brushing effect analysis results and the corresponding operating parameter adjustment history. The brushing effect analysis results include brushing force data, coverage integrity data, and posture error data. The operating parameter adjustment history includes vibration frequency adjustment information and amplitude adjustment information. Based on the historical data within a preset period, a brushing habit profile of the corresponding user is generated through a clustering algorithm. The profile classification is determined based on brushing force characteristics, coverage characteristics, and posture characteristics.

9. The electric toothbrush system based on artificial intelligence effect analysis according to claim 8, characterized in that, In the user interaction feedback module, when the brushing effect analysis results indicate that there is excessive brushing force, missed areas, or incorrect posture, the motor of the electric toothbrush is controlled to generate a vibration waveform with a sudden drop in intensity or a regular intermittent vibration waveform to provide real-time tactile warnings. After brushing, a visual report is generated, which includes brushing time, oral cavity zone coverage heatmap, comprehensive effect score, and a comparison of recent brushing effect trends, and is displayed to the user through the bound mobile terminal application.

Images