An Artificial Intelligence-Based Method and System for Bicycle Fatigue Detection

Abstract

This invention belongs to the field of data processing technology. It discloses an artificial intelligence-based method and system for bicycle fatigue detection, comprising: a data establishment end, an intelligent analysis end, and a safety intervention end. The data establishment end is used to collect brake and vibration datasets, providing accurate raw data for fatigue assessment. The intelligent analysis end is used to extract key features that accurately reflect fatigue status from preprocessed vibration and brake data. The safety intervention end is used to enable users to quickly perceive risks through multiple methods and proactively intervene in different potential hazards to reduce danger. This invention, based on artificial intelligence technology, accurately detects and assesses the fatigue status of the bicycle through multi-dimensional data collection and processing, and combines graded early warning, proactive intervention, and vehicle network linkage mechanisms to comprehensively ensure riding safety.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and more specifically, to an artificial intelligence-based method and system for detecting bicycle fatigue. Background Technology

[0002] As a popular means of transportation and exercise, bicycles are prone to fatigue wear on key components such as frames and brakes due to long-term vibration and friction. If not detected in time, this can lead to safety accidents such as breakage or failure. Bicycle fatigue detection systems can capture changes in component status in real time, identify potential hazards in advance, and provide risk warnings for riders. This not only avoids accidents caused by fatigue failures but also guides targeted maintenance, extends the service life of the vehicle, and meets the safety requirements of green travel scenarios.

[0003] Current bicycle fatigue detection technology has basic data collection capabilities, but there are still obvious limitations. Most solutions focus on detecting single components and ignore the fatigue transmission correlation between components. The detection logic has poor adaptability to riding conditions and is easily affected by road surface interference, leading to misjudgments. At the same time, the detection results mostly rely on manual analysis and lack intelligent quantitative evaluation capabilities, making it difficult to quickly output accurate and actionable safety recommendations and failing to fully meet the real-time protection needs in actual riding.

[0004] In view of this, the present invention proposes an artificial intelligence-based bicycle fatigue detection method and system to solve the above problems. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art and to achieve the above objectives, the present invention provides the following technical solution: the artificial intelligence-based bicycle fatigue detection system includes: a data establishment end, an intelligent analysis end, and a safety intervention end; The data establishment terminal includes a data acquisition module and a data preprocessing module; The data acquisition module includes a vibration data acquisition unit and a brake data acquisition unit; The vibration data acquisition unit is used to acquire vibration datasets, which include three-dimensional vibration characteristic data, vibration attenuation data, and vibration strain data. The brake data acquisition unit is used to collect brake datasets, which include brake pressure characteristic data, thermal fatigue data, and brake performance decay data. The data acquisition window length is dynamically adjusted based on riding speed. When the riding speed is greater than 25 km / h, the sampling period is 2, shortening the window to capture instantaneous changes. When the riding speed is greater than or equal to 10 km / h and less than or equal to 25 km / h, the sampling period is 5, balancing real-time performance and comprehensiveness with a medium window. When the riding speed is less than 10 km / h, the sampling period is 8, extending the window to obtain comprehensive data. The data preprocessing module includes an adaptive noise filtering unit and a data synchronization unit; The adaptive noise filtering unit is used to identify the current riding state based on the collected vibration and braking datasets, and outputs state labels including smooth riding, bumpy riding, normal braking, and emergency braking. Based on the output state labels, the core noise features under the current state label are first extracted, including the cutoff frequency and filter intensity. Then, specific filter parameters are generated through preset feature-parameter mapping rules. In the smooth riding state, the filter retains the frame's own vibration features greater than or equal to 8 Hz, and the filter intensity is set to level 1 for light filtering to avoid damaging the effective signal. In the bumpy riding state, an adaptive bandpass filter is used, with the passband range set to the current dominant vibration frequency, passing through the first two... The data collected in seconds shows that the filter intensity is set to level 3, medium filtering, suppressing impact noise while preserving the chassis strain characteristics; under emergency braking conditions, a combined filtering mode is adopted, with a high-pass filter cutoff frequency of 15 Hz to filter low-frequency impact noise, and adaptive noise reduction to track the screaming noise frequency in real time, dynamically adjusting the band-stop range, and the filter intensity is set to level 5 for high-intensity filtering to ensure clear braking performance data; under normal braking conditions, the band-stop filter center frequency is set to 10 Hz, and the filter intensity increases with increasing brake pressure. When the brake pressure characteristic data is less than or equal to 200N, the filter intensity is set to level 2, and when the brake pressure characteristic data is greater than 200N, the filter intensity is set to level 3. The data synchronization unit is used to time-align the remaining sensor data with the timestamp of the frame center shaft vibration sensor as a reference and the quadratic interpolation method, with the synchronization error controlled within 10 milliseconds, to ensure the correlation of data from different parts at the same time; it cleans all sub-data items in the basic dataset by filling missing values ​​and normalizes all sub-data items in the basic dataset to the range of [0, 1] based on the normalization formula; The intelligent analysis terminal includes a feature extraction module and an intelligent fatigue detection module; The feature extraction module includes a fatigue feature extraction unit and a feature filtering unit; The fatigue feature extraction unit is used to extract vibration fatigue features and braking fatigue features. Vibration fatigue features are extracted in both the time and frequency domains. The core time-domain feature is the peak factor, obtained by dividing the vibration peak value by the effective vibration value, directly reflecting the fatigue accumulation state of the frame. The core frequency-domain feature is the proportion of low-frequency energy, extracted by Fast Fourier Transform (FFT). A normal frame has a main frequency of 12 Hz, while a fatigued frame, due to structural weakening, will shift to a lower frequency range greater than or equal to 2 Hz (below 10 Hz), directly determined by the peak position in the frequency spectrum. Braking fatigue features are mechanical and thermal. The core mechanical feature is the pressure transmission efficiency, obtained by dividing the brake caliper clamping force by the lever pressure, reflecting braking force loss, which is less than or equal to 2.5 during fatigue. The core thermal feature is the thermal cycling coefficient, obtained by dividing the highest brake disc temperature by the cooling time. Based on this, a coupling correlation feature is obtained, reflecting the transmission relationship between braking action and frame vibration: in a normal vehicle, after the brake lever is subjected to force, the frame will... The vibration peak occurs within seconds, and the correlation between the two is greater than or equal to 0.8; when the brakes or frame are fatigued, the force transmission lags to greater than or equal to 0.3 seconds, and the correlation drops to below 0.5, which can accurately capture the fatigue transmission effect between components; The feature filtering unit is used to filter 10 core features, including brake coupling efficiency, chassis impact risk, brake mechanical fatigue, thermal cycling coefficient, vibration attenuation data, vibration strain data, brake performance attenuation data, chassis frequency offset, brake pressure fluctuation, and overall fatigue synergy value. The intelligent fatigue detection module includes a fatigue detection unit and a dynamic fatigue assessment unit; The fatigue detection unit is used to establish an intelligent detection model based on the transformer model, introducing two factors: state and fatigue. Furthermore, based on the transformer model, the steps for establishing an intelligent detection model by introducing a dual factor of state and fatigue include: Step 1: Use the 10-dimensional core features as input data and the state label as the 11th dimension feature, then fuse and concatenate them with the 10-dimensional core features to bind the state and fatigue data; simultaneously, determine the time window length based on cycling speed; for example, when the cycling speed is 20 kilometers per hour, the window length is: The input format is [32,6,10]; Step 2: The CNN layer is designed as a 2-layer one-dimensional convolutional neural network. The first convolutional layer has 11 input channels, a kernel size of 3×1, 16 output channels, and a stride of 1; the second convolutional layer has 16 input channels, a kernel size of 3×1, 32 output channels, and a stride of 1; the final output feature dimension is [32, 6, 32]. The ReLU activation function is used after convolution to achieve non-linear mapping of features and avoid the gradient vanishing problem. Step 3: Based on the transformer model, a state label weight matrix is ​​introduced, enabling the model to dynamically increase the attention of key features according to the current state. Under bumpy conditions, the weight of vibration features is amplified, and under braking conditions, the weight of coupling features is amplified. The [32,6,32] features output by the transformer are compressed to [32,32] dimensions through global average pooling and connected to two fully connected layers (64 hidden layer neurons and 1 output layer neuron). Finally, the fatigue probability in the 0-1 interval is output, completing the mapping from features to probability. Step 4: Introduce 6000 sets of samples, which are divided into a training set (4200 sets), a validation set (1200 sets), and a test set (600 sets) in a 7:2:1 ratio. Data annotation is completed by three mechanical engineers, with an annotation consistency of greater than or equal to 98%. The Adam optimizer is used, with an initial learning rate set to 0.001 and a step-decay strategy. The training cycle is set to a maximum of 50 rounds. If the validation set loss does not decrease for 5 consecutive rounds, training is terminated to avoid overfitting and obtain the intelligent detection model. The dynamic fatigue assessment unit, based on the fatigue probability output by the intelligent detection model, introduces aging coefficient and state loss coefficient for correction to obtain the final fatigue probability. The security intervention terminal includes an early warning intervention module and a system protection module; The early warning and intervention module includes a tiered early warning unit and a safety intervention unit; The graded early warning unit is used to design a four-level early warning system based on the final fatigue probability output by the intelligent fatigue detection module. When the final fatigue probability is less than or equal to 0.2, it is determined that there is no fatigue, the handlebar indicator light is solid green, and the mobile terminal displays that the vehicle is in good condition. When the final fatigue probability is greater than 0.2 and less than or equal to 0.4, it is determined that there is mild fatigue, the handlebar indicator light flashes green, and the mobile terminal displays that the vehicle is in mild fatigue. When the final fatigue probability is greater than 0.4 and less than or equal to 0.7, it is determined that there is moderate fatigue, the handlebar indicator light is solid yellow, the mobile terminal prompts that there is moderate fatigue, and the rider should pay attention to braking control and reduce fast riding. When the final fatigue probability is greater than 0.7, it is determined that there is severe fatigue, the handlebar indicator light is solid red, the mobile terminal pushes an emergency warning, and automatically locates the nearest bicycle repair shop. The safety intervention unit is used to perform active intervention based on key fatigue points through electronic control; when the final fatigue probability is greater than 0.4, the electronic brake damping increases and the mobile terminal displays the wear degree of the brake pads; when the final fatigue probability is greater than 0.6, the speed is limited by automatically adjusting the combination of the front and rear sprockets; if the brakes suddenly fail, the taillights will immediately flash at a frequency of 5 Hz, triggering a helmet vibration warning. The system support module includes a multi-scenario adaptation support unit and a self-calibration and diagnosis unit; The multi-scenario adaptation and protection unit is used for multi-model adaptation and all-environment adaptation. It features a modular bracket and dual Bluetooth interfaces, allowing for switching between mountain bikes and road bikes in 3 minutes. It also supports IP67 waterproof rating and operation from -10 degrees Celsius to 50 degrees Celsius. Furthermore, it enters sleep mode after 5 minutes of inactivity, reducing power consumption by 90%. It wakes up when vibrating and supports solar-assisted power supply for up to 60 days of battery life. The self-calibration diagnostic unit uses 5-second static reference data from a new vehicle of the same model as a standard and stores it in the cloud. Each time it starts, it collects 5 seconds of static data from the vehicle and calculates calibration coefficients to correct sensor errors. At the same time, it monitors the sensor communication status in real time. When the sensor has no data transmission for more than 5 seconds, it performs fault self-diagnosis, including cleaning the sensor probe, checking whether the sensor wiring is loose, and replacing the sensor battery. Based on this, it automatically saves the collected vibration dataset, brake dataset, calibration coefficients, and final fatigue probability to the cloud.

[0006] The technical effects and advantages of the artificial intelligence-based bicycle fatigue detection method and system of this invention are as follows: 1. This invention establishes a data acquisition terminal to collect vibration, pressure, and other data from key components through distributed acquisition, thereby constructing a comprehensive data system to obtain raw information. Based on this, it first combines status labels to dynamically filter interference, and then simultaneously calibrates multi-source data and enhances sample quality, forming a dual guarantee to effectively eliminate redundant noise and control the data interference rate at a low level, providing accurate and reliable basic support for subsequent fatigue assessment work.

[0007] 2. This invention sets up an intelligent analysis terminal to extract multi-dimensional features from preprocessed data, and retains core information after screening and optimization to form a more targeted analysis basis. Based on these high-value features, a model is built in combination with the actual situation to accurately capture potential fatigue signs. The intelligent algorithm outputs reliable fatigue assessment results, providing scientific and accurate decision support for subsequent safety warnings and intervention measures, and improving the effectiveness and credibility of the overall detection.

[0008] 3. This invention activates corresponding warnings based on fatigue assessment results by setting up a safety intervention terminal, and actively intervenes in potential hazards of key components; it is adaptable to different usage scenarios and vehicle models, reduces energy consumption through intelligent control, automatically calibrates each time it starts to ensure data reliability, and pushes troubleshooting guidance and backs up information when a fault occurs, which not only enhances riding safety, but also reduces maintenance trouble and cost, achieving the dual goals of safety assurance and system durability. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of the overall architecture of an artificial intelligence-based bicycle fatigue detection system according to the present invention; Figure 2This is a schematic diagram of the data establishment terminal architecture of an artificial intelligence-based bicycle fatigue detection system according to the present invention. Figure 3 This is a schematic diagram of the intelligent analysis terminal architecture of an artificial intelligence-based bicycle fatigue detection system according to the present invention. Figure 4 This is a schematic diagram of the safety intervention terminal architecture of an artificial intelligence-based bicycle fatigue detection system according to the present invention. Figure 5 This is a schematic diagram of the overall architecture of an artificial intelligence-based bicycle fatigue detection method according to the present invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0012] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to detection.” Similarly, depending on the context, the phrases “if determination” or “if detection (of the stated condition or event)” can be interpreted as “when determination” or “in response to determination” or “when detection (of the stated condition or event)” or “in response to detection (of the stated condition or event).”

[0013] Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0014] Example 1, please refer to Figures 1 to 2 As shown in the figure, the artificial intelligence-based bicycle fatigue detection system described in this embodiment includes: a data establishment end, an intelligent analysis end, and a safety intervention end; The data acquisition unit is used to collect brake and vibration datasets, providing accurate original data for fatigue assessment. For mechanical noise and asynchronous issues in the collected data, dynamic optimization processing is combined with riding status. Through noise filtering, time calibration, and data augmentation, invalid interference is eliminated, data format is standardized, and data quality and utilization are improved to ensure that the data can accurately reflect the actual condition of the vehicle. The intelligent analysis terminal is used to extract key features that accurately reflect fatigue status from preprocessed vibration and braking data. It filters out useless information, weights and highlights core safety-related features, and integrates them to form effective comprehensive features. Based on the comprehensive features, an intelligent model adapted to the riding scenario is used to quickly determine the fatigue status. The fatigue level is dynamically obtained by combining the vehicle usage, providing accurate and reliable decision-making basis for subsequent graded early warning and safety intervention. The safety intervention terminal is used to enable users to quickly perceive risks through various means, and to proactively intervene in different hidden dangers to reduce dangers; at the same time, it is compatible with a variety of usage scenarios and vehicle models, and its low power consumption design extends the battery life. It can also automatically calibrate, troubleshoot and back up data to ensure system reliability and reduce maintenance costs. The data creation end includes a data acquisition module and a data preprocessing module; The data acquisition module includes a vibration data acquisition unit and a brake data acquisition unit; The vibration data acquisition unit is used to acquire vibration datasets, which include three-dimensional vibration characteristic data, vibration attenuation data, and vibration strain data. It should be explained that, by installing vibration acceleration sensors at three key load-bearing points—the seat tube, bottom bracket, and rear fork—instantaneous vibration acceleration, frequency, and amplitude are collected at each load-bearing point during bicycle operation to obtain three-dimensional vibration characteristic data. This data directly reflects the degree of fatigue accumulation in the frame structure; as fatigue intensifies, the vibration frequency will show abnormal shifts. Vibration sensors are also used to collect the vibration waveform and vibration energy changes during the rotation of the hub, obtaining vibration attenuation data. Finally, vibration sensors are used to collect the strain vibration signal at the connection between the handlebar seat tube and the frame, obtaining vibration strain data. The brake data acquisition unit is used to collect brake datasets, which include brake pressure characteristic data, thermal fatigue data, and brake performance decay data. To achieve a balance between real-time performance and comprehensiveness in data acquisition, the data acquisition module dynamically adjusts the data acquisition window length based on riding speed. When the riding speed is greater than 25 km / h, the sampling period is 2, shortening the window to capture instantaneous changes. When the riding speed is greater than or equal to 10 km / h and less than or equal to 25 km / h, the sampling period is 5, using a medium window to balance real-time performance and comprehensiveness. When the riding speed is less than 10 km / h, the sampling period is 8, extending the window to obtain comprehensive data. It should be explained that, through pressure sensors, the lever pressure, brake caliper clamping force, and pressure change rate during the braking operation of a specified bicycle are collected to obtain brake pressure characteristic data; through temperature sensors, the temperature change curve and maximum temperature value of the brake disc of a specified bicycle during the braking process are collected to obtain thermal fatigue data; through speed sensors, the rate of decrease in wheel speed and braking response time of a specified bicycle during the braking process are collected to obtain braking performance decay data. The data preprocessing module includes an adaptive noise filtering unit and a data synchronization unit; The adaptive noise filtering unit identifies the current riding state based on the collected vibration and braking datasets, outputting state labels including smooth riding, bumpy riding, normal braking, and emergency braking. Based on the output state labels, it first extracts the core noise features under the current state label, including the cutoff frequency and filter intensity, and then generates specific filter parameters through preset feature-parameter mapping rules. In the smooth riding state, the filter retains the frame's own vibration features greater than or equal to 8 Hz, with the filter intensity set to level 1 for light filtering to avoid damaging the effective signal. In the bumpy riding state, an adaptive bandpass filter is used, with the passband range set to the current dominant vibration frequency, passing through the first 2 seconds. The collected data shows that the filter intensity is set to level 3 for medium filtering, suppressing impact noise while preserving the chassis strain characteristics. Under emergency braking conditions, a combined filtering mode is used, with a high-pass filter cutoff frequency of 15 Hz to filter low-frequency impact noise and adaptive noise reduction to track the screaming noise frequency in real time, dynamically adjusting the band-stop range. The filter intensity is set to level 5 for high-intensity filtering to ensure clear braking performance data. Under normal braking conditions, the band-stop filter center frequency is set to 10 Hz, and the filter intensity increases with increasing brake pressure. When the brake pressure characteristic data is less than or equal to 200N, the filter intensity is set to level 2; when the brake pressure characteristic data is greater than 200N, the filter intensity is set to level 3. It should be explained that when the brake pressure is less than or equal to 50N, the frame vibration frequency is less than or equal to 5Hz, and the vibration energy percentage is less than or equal to 10%, it is considered a smooth ride; when the frame vibration frequency is between 15 and 50Hz and the vibration energy percentage is greater than or equal to 30%, it is considered a bumpy ride; when the brake pressure change rate is between 5 and 30N / s and the vibration frequency is between 8 and 12Hz, it is considered a normal braking; when the brake pressure change rate is greater than 30N / s, the vibration peak value is greater than 10 meters per square second, and the wheel speed decay rate is greater than 5 meters per square second, it is considered an emergency braking. The data synchronization unit is used to time-align the remaining sensor data with the timestamp of the frame center axle vibration sensor as a reference and the quadratic interpolation method. The synchronization error is controlled within 10 milliseconds to ensure the correlation of data from different parts at the same time. The missing value filling is used to clean all the sub-data items in the basic dataset and normalizes all the sub-data items in the basic dataset to the range of [0, 1] based on the normalization formula. It needs to be explained that the basic dataset includes the vibration dataset and the braking dataset; missing value imputation refers to filling in missing data by using the mean of the previous and next nearest values ​​of the missing data when a certain data item is missing; the specific expression of the normalization formula is as follows: The normalized value is obtained. ,in, Any sub-data item of the basic data, This represents the historical maximum value of any given sub-data item. This represents the historical minimum value of any given sub-data item; normalization is used to eliminate the dimensions of all sub-data items in the base dataset. Example 2, please refer to Figure 3 As shown: Based on Embodiment 1, the intelligent analysis terminal includes a feature extraction module and an intelligent fatigue detection module; The feature extraction module includes a fatigue feature extraction unit and a feature filtering unit; The fatigue feature extraction unit is used to extract vibration fatigue features and brake fatigue features. Vibration fatigue features are extracted in both the time and frequency domains. The core time-domain feature is the peak factor, obtained by dividing the vibration peak value by the effective vibration value, directly reflecting the fatigue accumulation state of the frame. The core frequency-domain feature is the proportion of low-frequency energy, extracted by Fast Fourier Transform (FFT). A normal frame's main frequency is 12 Hz, while a fatigued frame, due to structural weakening, will shift to a lower frequency range greater than or equal to 2 Hz (below 10 Hz), directly determined by the peak position in the frequency spectrum. Brake fatigue features are mechanical and thermal. The core mechanical feature is the pressure transmission efficiency, obtained by dividing the brake caliper clamping force by the lever pressure, reflecting brake force loss; it is less than or equal to 2.5 during fatigue. The core thermal feature is the thermal cycling coefficient, obtained by dividing the highest brake disc temperature by the cooling time. Based on this, a coupling correlation feature is obtained, reflecting the transmission relationship between braking action and frame vibration: in a normal vehicle, after the brake lever is subjected to force, the frame will... The vibration peak occurs within seconds, and the correlation between the two is greater than or equal to 0.8; when the brakes or frame are fatigued, the force transmission lags to greater than or equal to 0.3 seconds, and the correlation drops to below 0.5, which can accurately capture the fatigue transmission effect between components; It should be explained that the vibration peak value is the maximum vibration value in the collected vibration data, and the vibration effective value is a statistic obtained by first squaring the collected vibration acceleration time domain data, then taking the average value, and finally taking the square root. The feature filtering unit is used to filter 10 core features, including brake coupling efficiency, chassis impact risk, brake mechanical fatigue, thermal cycling coefficient, vibration attenuation data, vibration strain data, brake performance attenuation data, chassis frequency offset, brake pressure fluctuation, and overall fatigue synergy value. It should be explained that the braking coupling efficiency is 0.6 times the braking response delay; the frame impact risk is the sum of the vibration peak value and the peak factor of 0.5 times; the braking mechanical fatigue is 3 minus the pressure change rate; the frame frequency offset is 12 minus the current dominant vibration frequency; and the overall fatigue synergy value is obtained by dividing the weighted sum of the first 9 features by 9. The intelligent fatigue detection module includes a fatigue detection unit and a dynamic fatigue assessment unit; The fatigue detection unit is used to establish an intelligent detection model based on the transformer model, incorporating both state and fatigue factors. Furthermore, based on the transformer model, the steps to establish an intelligent detection model by introducing a dual factor of state and fatigue include: Step 1: Use the 10-dimensional core features as input data and the state label as the 11th dimension feature, then fuse and concatenate them with the 10-dimensional core features to bind the state and fatigue data; simultaneously, determine the time window length based on cycling speed; for example, when the cycling speed is 20 kilometers per hour, the window length is: The input format is [32,6,10]; It should be explained that the input format is 32 for batch size, 6 for number of time slices, and 10 for feature data; Step 2: The CNN layer is designed as a 2-layer one-dimensional convolutional neural network. The first convolutional layer has 11 input channels (10-dimensional fused features and 1-dimensional state label), a kernel size of 3×1, 16 output channels, and a stride of 1. The second convolutional layer has 16 input channels, a kernel size of 3×1, 32 output channels, and a stride of 1. The final output feature dimension is [32, 6, 32] (batch size × number of time slices × number of feature channels). The ReLU activation function is used after convolution to achieve non-linear mapping of features and avoid the gradient vanishing problem. Step 3: Based on the transformer model, a state label weight matrix is ​​introduced, enabling the model to dynamically increase the attention of key features according to the current state. Under bumpy conditions, the weight of vibration features is amplified, and under braking conditions, the weight of coupling features is amplified. The [32,6,32] features output by the transformer are compressed to [32,32] dimensions through global average pooling and connected to two fully connected layers (64 hidden layer neurons and 1 output layer neuron). Finally, the fatigue probability in the 0-1 interval is output, completing the mapping from features to probability. Step 4: Introduce 6000 sets of samples, divided into a training set (4200 sets), a validation set (1200 sets), and a test set (600 sets) in a 7:2:1 ratio. Data annotation was completed by three mechanical engineers, with an annotation consistency of greater than or equal to 98%. The Adam optimizer was used, with an initial learning rate set to 0.001 and a step-decay strategy (decaying to 80% of the current value every 10 training epochs). The maximum number of training epochs was set to 50. Training was terminated if the validation set loss did not decrease for 5 consecutive epochs to avoid overfitting and obtain the intelligent detection model. The dynamic fatigue assessment unit, based on the fatigue probability output by the intelligent detection model, incorporates aging and state loss coefficients for correction to obtain the final fatigue probability. The specific calculation formula is as follows: To obtain the final fatigue probability ,in, The fatigue probability output by the intelligent detection model. The aging coefficient is set according to the vehicle's service life. For vehicles less than 1 year old, k1 is 0.9 (new cars have strong fatigue resistance), for 1-3 years old, k1 is 1.0 (performance stable period), and for more than 3 years old, k1 is 1.1 (structural aging accelerates fatigue). At the same time, a component material correction coefficient is introduced: 1.0 for steel frames, 0.8 for aluminum alloys, and 0.6 for carbon fiber. Therefore, the aging coefficient k is K1 multiplied by the component material correction coefficient. The condition loss coefficient is set according to the intensity of riding condition loss. It is 1.0 for smooth riding, 1.2 for bumpy riding (road impact aggravates fatigue), 1.3 for normal braking (frequent mechanical wear), and 1.5 for emergency braking (large instantaneous load, significant loss). Example 3, please refer to Figure 4 As shown: Based on Embodiment 1, the security intervention terminal includes an early warning intervention module and a system protection module; The early warning and intervention module includes a tiered early warning unit and a safety intervention unit; The graded early warning unit is designed with four levels of warning based on the final fatigue probability output by the intelligent fatigue detection module. When the final fatigue probability is less than or equal to 0.2, it is judged as no fatigue, the handlebar indicator light is solid green, and the mobile terminal displays that the vehicle is in good condition. When the final fatigue probability is greater than 0.2 and less than or equal to 0.4, it is judged as mild fatigue, the handlebar indicator light flashes green, and the mobile terminal displays that the vehicle is in mild fatigue. When the final fatigue probability is greater than 0.4 and less than or equal to 0.7, it is judged as moderate fatigue, the handlebar indicator light is solid yellow, the mobile terminal prompts moderate fatigue, and advises attention to brake control and reduction of fast riding. When the final fatigue probability is greater than 0.7, it is judged as severe fatigue, the handlebar indicator light is solid red, the mobile terminal pushes an emergency warning, and automatically locates the nearest bicycle repair shop. The safety intervention unit is used to perform active intervention based on key fatigue points through electronic control; when the final fatigue probability is greater than 0.4, the electronic brake damping increases and the wear degree of the brake pads is displayed on the mobile terminal; when the final fatigue probability is greater than 0.6, the speed is limited by automatically adjusting the combination of the front and rear sprockets; if the brakes suddenly fail, the taillights will immediately flash at a frequency of 5 Hz, triggering a helmet vibration warning. The system support module includes a multi-scenario adaptation support unit and a self-calibration diagnostic unit; The multi-scenario adaptation and protection unit is used for multi-model adaptation and all-environment adaptability. The modular bracket and Bluetooth dual interface can complete the switching between mountain bike and road bike in 3 minutes. It also supports IP67 waterproof rating and can work from -10 degrees Celsius to 50 degrees Celsius. At the same time, it enters sleep mode after 5 minutes of inactivity to reduce power consumption by 90%. It can be woken up by vibration and supports solar-assisted power supply with a battery life of 60 days. The self-calibration diagnostic unit uses 5-second static reference data from a new vehicle of the same model as a standard, which is stored in the cloud. Each time it starts, it collects 5 seconds of static data from the vehicle itself, calculates calibration coefficients to correct sensor errors, and monitors sensor communication status in real time. When the sensor has no data transmission for more than 5 seconds, it performs fault self-diagnosis, including cleaning the sensor probe (removing brake dust), checking whether the sensor wiring is loose, and replacing the sensor battery. Based on this, it automatically saves the collected vibration dataset, brake dataset, calibration coefficients, and final fatigue probability to the cloud. It should be explained that the calibration factor is the reference data divided by the measured value; The beneficial effects of this embodiment are that by accurately processing riding data, it effectively eliminates interfering information and ensures reliable judgment of vehicle status; it can keenly detect potential fatigue signs of components, output accurate assessments based on actual usage, provide early warnings of risks, and coordinate with surrounding areas to form protection; it can proactively intervene in potential hazards to prevent escalation of danger; the system is compatible with various vehicle models, intelligently controls and reduces energy consumption, automatically calibrates each time it starts to ensure data accuracy, and pushes troubleshooting guidance and backs up information when a fault occurs, thus both strengthening the safety defense line and reducing maintenance costs.

[0015] Example 4, please refer to Figure 5 As shown, the parts not described in detail in this embodiment are described in Embodiments 1 to 3. An artificial intelligence-based bicycle fatigue detection method is provided. The method includes: S1: collecting vibration, pressure, and temperature data of the frame and brake parts through a distributed sensor array to construct a three-dimensional system; S2: Identify the state dynamic filtering, then synchronously calibrate multi-source data and enhance samples, providing double protection to remove noise and reduce data interference rate; S3: Extract vibration, braking, and time-domain and frequency-domain features, and after screening, weighting and fusion, obtain 10-dimensional core features; S4: Introduce state weights to build an intelligent diagnostic model. By extracting features, train a lightweight model and combine it with the aging coefficient to output accurate fatigue assessment. S5: Activate graded early warning according to fatigue level, link vehicle network to warn the surrounding area, and perform active intervention for brake and frame fatigue; S6: Adaptable to multiple vehicle models, intelligently controls power consumption, automatically calibrates sensors upon startup, and pushes troubleshooting methods and backs up data to the cloud when a fault occurs.

[0016] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the present invention.

Claims

1. A bicycle fatigue detection system based on artificial intelligence, characterized in that, The system includes: a data creation terminal, an intelligent analysis terminal, and a security intervention terminal; The data establishment terminal is used to collect brake datasets and vibration datasets. For mechanical noise issues in the collected data, combined with the dynamic optimization processing method based on riding status, invalid interference is eliminated through noise filtering and data enhancement. The intelligent analysis terminal is used to extract key features reflecting fatigue status from preprocessed vibration and braking data. By filtering out useless information and highlighting core safety-related features, it uses an intelligent diagnostic model adapted to riding scenarios to quickly determine fatigue status and dynamically obtain fatigue level by combining vehicle usage time. The safety intervention terminal is used to enable users to quickly perceive risks through various means and proactively intervene in different hidden dangers; at the same time, it is compatible with various usage scenarios and vehicle models, and its low power consumption design extends battery life, and it automatically calibrates sensors, troubleshoots faults, and backs up data; The intelligent analysis terminal includes a feature extraction module and an intelligent fatigue detection module; The feature extraction module includes a fatigue feature extraction unit and a feature filtering unit; The fatigue feature extraction unit is used to extract vibration fatigue features and braking fatigue features. Vibration fatigue features are extracted in both time and frequency domains. Braking fatigue features are extracted in both mechanical and thermal domains, and coupled correlation features are obtained to reflect the transmission relationship between braking action and frame vibration. The intelligent fatigue detection module includes a fatigue detection unit and a dynamic fatigue assessment unit; The fatigue detection unit first uses 10-dimensional core features and state labels as 11-dimensional input tensors, and dynamically adjusts the data window length according to the riding speed to complete the adaptation; then, a two-layer convolutional neural network extracts local fatigue features, and then a Transformer that incorporates the state weight matrix dynamically improves the attention of key features to achieve dual-factor feature fusion; finally, the fatigue probability is output through a fully connected layer.

2. The bicycle fatigue detection system based on artificial intelligence according to claim 1, characterized in that, The system also includes: the data establishment terminal includes a data acquisition module and a data preprocessing module; The data acquisition module includes a vibration data acquisition unit and a brake data acquisition unit; The vibration data acquisition unit is used to acquire vibration datasets, which include three-dimensional vibration characteristic data, vibration attenuation data, and vibration strain data. The brake data acquisition unit is used to acquire brake datasets, which include brake pressure characteristic data, thermal fatigue data, and brake performance decay data. Simultaneously, the data acquisition window length is dynamically adjusted based on riding speed. When the riding speed is greater than 25 km / h, the sampling period is 2 cycles; when the riding speed is greater than or equal to 10 km / h and less than or equal to 25 km / h, the sampling period is 5 cycles; and when the riding speed is less than 10 km / h, the sampling period is 8 cycles.

3. The bicycle fatigue detection system based on artificial intelligence according to claim 2, characterized in that, The data preprocessing module includes an adaptive noise filtering unit and a data synchronization unit; The adaptive noise filtering unit is used to identify the current riding state based on the collected vibration dataset and brake dataset, and output a state label. Based on the output state label, the core noise features under the current state label are first extracted, and then exclusive filtering parameters are generated through preset feature and parameter mapping rules.

4. The bicycle fatigue detection system based on artificial intelligence according to claim 3, characterized in that, The data synchronization unit is used to time-align the remaining sensor data using a quadratic interpolation method based on the timestamp of the frame axle vibration sensor; it cleans all sub-data items in the basic dataset by filling in missing values, and normalizes all sub-data items in the basic dataset to the range of [0, 1] based on the normalization formula.

5. The bicycle fatigue detection system based on artificial intelligence according to claim 1, characterized in that, The feature filtering unit is used to filter 10 core features, including brake coupling efficiency, chassis impact risk, brake mechanical fatigue, thermal cycling coefficient, vibration attenuation data, vibration strain data, brake performance attenuation data, chassis frequency offset, brake pressure fluctuation, and overall fatigue synergy value.

6. The bicycle fatigue detection system based on artificial intelligence according to claim 1, characterized in that, The dynamic fatigue assessment unit, based on the fatigue probability output by the intelligent detection model, introduces aging coefficient and state loss coefficient for correction to obtain the final fatigue probability.

7. The bicycle fatigue detection system based on artificial intelligence according to claim 1, characterized in that, The security intervention terminal includes an early warning intervention module and a system protection module; The early warning and intervention module includes a tiered early warning unit and a safety intervention unit; The graded early warning unit is used to design a four-level early warning system based on the final fatigue probability output by the intelligent fatigue detection module. The safety intervention unit is used to perform active intervention based on key fatigue points through electronic control. When the final fatigue probability is greater than 0.4, the electronic brake damping increases, and the mobile terminal displays the degree of brake pad wear. When the final fatigue probability is greater than 0.6, the speed is limited by automatically adjusting the combination of the front and rear sprockets; if the brakes suddenly fail, the taillights will flash immediately at a frequency of 5 Hz, triggering a helmet vibration warning.

8. The bicycle fatigue detection system based on artificial intelligence according to claim 7, characterized in that, The system support module includes a multi-scenario adaptation support unit and a self-calibration and diagnosis unit; The multi-scenario adaptation and protection unit is used for multi-model adaptation and all-environment adaptation. It features a modular bracket and dual Bluetooth interfaces, allowing for switching between mountain bikes and road bikes in 3 minutes. It also supports IP67 waterproof rating and operation from -10 degrees Celsius to 50 degrees Celsius. Furthermore, it enters sleep mode after 5 minutes of inactivity, reducing power consumption by 90%. It wakes up when vibrating and supports solar-assisted power supply for up to 60 days of battery life. The self-calibration diagnostic unit uses 5-second static reference data from a new vehicle of the same model as a standard and stores it in the cloud. Each time it starts, it collects 5 seconds of static data from the vehicle and calculates calibration coefficients to correct sensor errors. At the same time, it monitors the sensor communication status in real time. When the sensor has no data transmission for more than 5 seconds, it performs fault self-diagnosis, including cleaning the sensor probe, checking whether the sensor wiring is loose, and replacing the sensor battery. It automatically saves the collected vibration dataset, brake dataset, calibration coefficients, and final fatigue probability to the cloud.

9. A method for detecting bicycle fatigue based on artificial intelligence, implemented according to any one of claims 1-8, characterized in that, The work includes the following steps: S1: Collect vibration, pressure, and temperature data of the chassis and brake components through a distributed sensor array to construct a three-dimensional system; S2: Identify the state dynamic filtering, then synchronously calibrate multi-source data and enhance samples, providing double protection to remove noise and reduce data interference rate; S3: Extract vibration, braking, and time-domain and frequency-domain features, and after screening, weighting and fusion, obtain 10-dimensional core features; S4: Introduce state weights to build an intelligent diagnostic model. By extracting features, train a lightweight model and combine it with the aging coefficient to output accurate fatigue assessment. S5: Activate graded early warning according to fatigue level, link vehicle network to warn the surrounding area, and perform active intervention for brake and frame fatigue; S6: Adaptable to multiple vehicle models, intelligently controls power consumption, automatically calibrates sensors upon startup, and pushes troubleshooting methods and backs up data to the cloud when a fault occurs.

Images