Vehicle braking system oil leakage early warning method based on LSTM

Abstract

The invention discloses a vehicle braking system oil leakage early warning method based on LSTM, and relates to the technical field of vehicle braking, and the method comprises the following steps: collecting pressure data of a braking main cylinder or a wheel cylinder through a pressure sensor of a vehicle braking system, and then forming original pressure time sequence data; de-noising and normalization processing are carried out on the original pressure time sequence data, a sliding window is used for segmenting the processed original pressure time sequence data into continuous pressure sequence windows, and each pressure sequence window comprises N sampling points; an LSTM auto-encoder model is constructed, a pressure sequence window of a normal working condition is used as training data, training is carried out by taking minimization of a reconstruction error as a target, and if the reconstruction error exceeds a threshold value and lasts for a preset number of windows, an oil leakage early warning signal is triggered; a triggered early warning signal is output through the vehicle-mounted communication system, and the effect is that the danger of brake failure caused by leakage expansion is avoided.

Description

Technical Field

[0001] This invention relates to the field of vehicle braking technology, and more specifically, to a method for early warning of fluid leakage in vehicle braking systems based on LSTM. Background Technology

[0002] Traditional brake fluid leak detection relies on level sensors, typically placing a level probe in the brake fluid reservoir. An alarm is triggered when the fluid level falls below a threshold due to leakage. However, this approach only responds after significant fluid loss has occurred, failing to provide early warnings. Some solutions use a simple pressure threshold method, pre-setting a fixed pressure range. When the brake system pressure exceeds this range, it's considered abnormal. However, in actual vehicle operation, brake pressure fluctuates due to driving conditions (such as heavy loads or downhill braking). This fixed threshold method is prone to false alarms. For example, frequent braking during downhill driving of a heavily loaded vehicle can cause a brief pressure increase, which might be misinterpreted as a leak, reducing the reliability of the warning system. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the purpose of this invention is to provide a method for early warning of oil leakage in vehicle braking system based on LSTM.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for early warning of fluid leakage in a vehicle braking system based on LSTM, the method comprising the following steps: The pressure data of the master cylinder or wheel cylinder is collected by the pressure sensor of the vehicle braking system to form the raw pressure timing data; The original pressure time series data is denoised and normalized, and a sliding window is used to divide the processed original pressure time series data into continuous pressure sequence windows, each pressure sequence window containing N sampling points. An LSTM autoencoder model is constructed, using a stress sequence window under normal operating conditions as training data. The training aims to minimize the reconstruction error, and an early warning threshold is set based on the reconstruction error distribution of the training set. Real-time pressure data is collected, preprocessed, and then input into a trained LSTM autoencoder model to calculate the reconstruction error; an oil leakage early warning signal is triggered based on the reconstruction error. If the reconstruction error exceeds the threshold and continues for a preset number of windows, an oil leak warning signal will be triggered. The warning signal is triggered by outputting through the vehicle communication system.

[0005] Preferably, an oil leakage early warning signal is triggered based on the reconstruction error, specifically as follows: If the reconstruction error exceeds the threshold and continues for a preset number of windows, an oil leak warning signal will be triggered.

[0006] Preferably, the warning signal triggered by the vehicle communication system includes the following steps: The warning signal is output through at least one of the following methods: CAN bus, vehicle terminal, instrument light, buzzer, or vehicle networking platform.

[0007] Preferably, the sampling frequency of the pressure sensor is 100Hz; The length of the sliding window is set to 50 sampling points, corresponding to a duration of 0.5 seconds.

[0008] Preferably, the original pressure time series data is denoised using a moving average or wavelet filtering method; The raw pressure time series data were normalized using Z-score or Min-Max methods.

[0009] Preferably, the hidden layer dimension of the LSTM autoencoder model is ≤64, and the inference latency of the LSTM autoencoder model is <10ms.

[0010] Preferably, the warning threshold is set as follows: Calculate the mean of the reconstruction error of normal training samples plus three times the standard deviation.

[0011] Preferably, the LSTM autoencoder model can be replaced by a gated loop unit (GRU) or a Transformer encoder.

[0012] Preferably, the basic unit of the LSTM autoencoder model includes a forget gate, an input gate, and an output gate structure; The activation control of the forget gate, input gate, and output gate is determined by... Function implementation; The input gate also includes a function for generating candidate cell states. layer; At each time step, the hidden state of the cell in the previous time step. Unit status and current input For input; The forget gate controls the state of historical units. The degree of retention is controlled by the input gate for the current candidate state. The update degree of ' is used to calculate the current cell state. ; The output gate depends on the current cell state. and hidden state Control the output of the current time step .

[0013] Preferably, the process further includes, before triggering the oil leak warning signal: Calculate the distance between the pressure data characteristics and the curves of different disturbance modes to determine whether the disturbance source is located inside the monitored pipe section; If the disturbance source is determined to be located within the pipe section, an oil leak warning signal will be triggered. If the disturbance source is determined to be outside the pipe section, it is identified as external equipment operation and will not trigger an oil leak warning signal.

[0014] Compared with the prior art, the present invention has the following beneficial effects: This invention collects time-series pressure data from the brake master cylinder or wheel cylinders using pressure sensors, then performs denoising and normalization to eliminate interference. A sliding window segmentation sequence is then used to make the data more closely resemble actual pressure changes. The subsequently constructed LSTM autoencoder model is trained only on normal operating condition data, using reconstruction error as the core criterion. An early warning is triggered when the error exceeds a threshold and continues for a preset number of windows, avoiding false alarms caused by accidental interference and making the warning results more reliable. This process of real-time pressure data acquisition, real-time preprocessing, and inputting the data into the model to calculate reconstruction error, compared to traditional methods relying on manual inspection or delayed alarms after a drop in fluid level, can detect abnormal pressure changes in the early stages of a leak, preventing the risk of brake failure due to leak expansion. Attached Figure Description

[0015] Figure 1 A schematic diagram illustrating the steps of an LSTM-based method for early warning of fluid leakage in a vehicle braking system, as proposed in this invention. Figure 2 This is a diagram of the LSTM autoencoder network structure in this embodiment; Figure 3 This is a system architecture diagram for this embodiment. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0018] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0019] Reference Figures 1-3 As shown.

[0020] The embodiments further illustrate the LSTM-based early warning method for vehicle braking system fluid leakage proposed in this invention.

[0021] A method for early warning of fluid leakage in a vehicle braking system based on LSTM, the method comprising the following steps: The pressure data of the master cylinder or wheel cylinder is collected by the pressure sensor of the vehicle braking system to form the raw pressure timing data; The original pressure time series data is denoised and normalized, and a sliding window is used to divide the processed original pressure time series data into continuous pressure sequence windows, each pressure sequence window containing N sampling points. An LSTM autoencoder model is constructed, using a stress sequence window under normal operating conditions as training data. The training aims to minimize the reconstruction error, and an early warning threshold is set based on the reconstruction error distribution of the training set. Real-time pressure data is collected, preprocessed, and then input into a trained LSTM autoencoder model to calculate the reconstruction error; an oil leakage early warning signal is triggered based on the reconstruction error. If the reconstruction error exceeds the threshold and continues for a preset number of windows, an oil leak warning signal will be triggered. The warning signal is triggered by outputting through the vehicle communication system.

[0022] The oil leakage early warning signal is triggered based on the reconstruction error, specifically as follows: If the reconstruction error exceeds the threshold and continues for a preset number of windows, an oil leak warning signal will be triggered.

[0023] The reconstruction error is the result obtained by processing the stress sequence window through the LSTM autoencoder model. The LSTM autoencoder model reconstructs the input stress sequence, and the difference between the actual input sequence and the sequence reconstructed by the model is the reconstruction error. The warning threshold is determined based on training data under normal operating conditions. Specifically, it is calculated by adding three times the standard deviation to the mean of the reconstruction error of normal training samples. This value represents the reasonable upper limit of the fluctuation of the reconstruction error under normal operating conditions.

[0024] During the real-time monitoring phase, pressure data from the braking system is continuously collected. This data is processed into corresponding pressure sequence windows and then input into the trained LSTM autoencoder model to obtain the real-time reconstruction error. At this point, a dual assessment of the reconstruction error is required: first, whether the error exceeds a preset warning threshold. If the error does not exceed the threshold, it indicates that the fluctuation of the current pressure sequence is within the normal range, and no warning will be triggered; second, whether the state of error exceeding the threshold can continue for a preset number of windows. A window refers to a pressure sequence unit after being divided by a sliding window. For example, if the sliding window length is set to 50 sampling points, corresponding to a duration of 0.5 seconds, and the preset number of windows is 3, it means that the state of reconstruction error exceeding the threshold needs to continue for 1.5 seconds.

[0025] Assuming a vehicle's braking system is in normal operation, the LSTM autoencoder model calculates a reconstruction error with a mean of 0.2 and a standard deviation of 0.05. The warning threshold is then 0.2 plus 3 multiplied by 0.05, which equals 0.35. If the reconstruction error for a pressure sequence window at a certain moment reaches 0.4, exceeding the 0.35 threshold, but this is only a single instance of exceeding the threshold, a warning will not be triggered immediately. If the reconstruction errors for the subsequent two consecutive pressure sequence windows are 0.38 and 0.42, respectively, both exceeding the threshold, and this exceeds the threshold for three consecutive windows, meeting the preset duration condition, then a fluid leak warning signal is triggered. This ensures the accuracy of the warning and avoids false alarms caused by a single, accidental fluctuation.

[0026] The warning signal triggered by the vehicle communication system includes the following steps: The warning signal is output through at least one of the following methods: CAN bus, vehicle terminal, instrument light, buzzer, or vehicle networking platform.

[0027] First, a warning signal is output via the CAN bus. The CAN bus is a communication network between various electronic devices inside the vehicle. When an oil leak warning is triggered, the warning signal is transmitted via the CAN bus to various relevant control modules of the vehicle, such as the body control module. In this way, different systems of the vehicle can simultaneously obtain the warning information and provide data support for subsequent auxiliary operations.

[0028] In-vehicle terminals are typically information display devices inside the vehicle, such as the central control screen. When a warning signal is triggered, the in-vehicle terminal displays the corresponding warning information on the screen, such as directly displaying a text message warning of oil leakage, allowing the driver to see the warning content intuitively and thus understand the vehicle's status in a timely manner.

[0029] The vehicle's dashboard is one of the areas that drivers pay close attention to at all times. When there is a fluid leak warning, the corresponding warning indicator light on the dashboard will light up, such as a dedicated brake system malfunction indicator light or a fluid leak warning light. This visual alert can quickly attract the driver's attention, and even if the driver is not looking at the central control screen, they can still detect the abnormality through the dashboard lights.

[0030] A buzzer is an audio alert device in a vehicle that emits a specific warning sound when a warning signal is triggered.

[0031] The vehicle-to-everything (V2X) platform is a network platform that connects vehicles with external management systems. Warning signals are transmitted to the corresponding platform server via the V2X, such as the vehicle's fleet management platform or the manufacturer's service platform. This allows vehicle managers or manufacturer service personnel to remotely obtain vehicle warning information and follow up in a timely manner, such as contacting the driver to understand the situation or arranging rescue services.

[0032] The pressure sensor has a sampling frequency of 100Hz; The sliding window is set to a length of 50 sampling points, corresponding to a duration of 0.5 seconds.

[0033] The pressure sensor's sampling frequency is set to 100Hz, meaning that the pressure sensor will collect pressure data from the master cylinder or wheel cylinder 100 times per second. This sampling frequency ensures that the system can acquire detailed pressure changes in a dense and timely manner, without missing rapid pressure fluctuations during braking, providing sufficiently detailed raw data support for subsequent analysis of the pressure's temporal characteristics.

[0034] The sliding window is set to a length of 50 sampling points. The function of the sliding window is to divide the continuous pressure time series data into continuous sequence units, each unit containing pressure change information within 0.5 seconds.

[0035] The original pressure time series data were denoised using moving average or wavelet filtering. The raw pressure time series data were normalized using Z-score or Min-Max methods.

[0036] Denoising is achieved using either moving average or wavelet filtering. Moving average calculates the average of several consecutive sampling points in the original pressure time series data, replacing the corresponding values ​​in the original sequence with this average, thus smoothing out random fluctuations in the data. If a sliding window of 5 sampling points is selected, when processing the 3rd sampling point, the pressure values ​​from the 1st to the 5th sampling points are averaged, and this average is used as the value of the 3rd sampling point after denoising, thus mitigating the random errors of single sampling. Wavelet filtering, on the other hand, decomposes the original pressure data into components of different frequencies using wavelet transform, filters out the frequency components corresponding to the effective pressure signal, removes high-frequency noise components, and then reconstructs the data.

[0037] Z-score or Min-Max normalization is performed. Z-score calculates the mean and standard deviation of the original pressure time series data, then subtracts the mean from the value of each sample point and divides by the standard deviation. For example, if the mean of a set of original pressure data is 2 MPa and the standard deviation is 0.5 MPa, and the pressure value of one sample point is 3 MPa, after Z-score processing, the value of that point becomes (3-2) / 0.5 = 2, thus eliminating the absolute difference in pressure values ​​under different operating conditions. Min-Max, on the other hand, finds the maximum and minimum values ​​of the original pressure data, subtracts the minimum value from the value of each sample point, and divides by the difference between the maximum and minimum values, mapping the data to the interval between 0 and 1. For example, if the maximum value of the original pressure data is 5 MPa and the minimum value is 1 MPa, and the pressure value of a sample point is 3 MPa, after Min-Max processing, the value of that point becomes (3-1) / (5-1) = 0.5, ensuring that pressure data of different magnitudes are within a uniform numerical range.

[0038] The hidden layer dimension of the LSTM autoencoder model is ≤64, and the inference latency of the LSTM autoencoder model is <10ms.

[0039] The hidden layer dimension of the LSTM autoencoder model is set to no more than 64. The hidden layer dimension represents the feature complexity that the model can capture during the learning process. Too high a dimension leads to redundant model parameters and a surge in computational cost, making it difficult to run on the limited computing power of automotive hardware. Conversely, too low a dimension prevents the model from fully learning the effective features in the pressure time-series data, affecting the accuracy of subsequent reconstruction error calculations. Keeping the hidden layer dimension below 64 allows the model to sufficiently learn the temporal variation patterns of braking pressure data, such as the pressure rise slope and stabilization time, while avoiding the computational burden caused by too many parameters. For example, when processing a pressure sequence window containing 50 sampling points, a 64-dimensional hidden layer can encode this time-series data into a vector containing 64 features, preserving the core information of pressure changes without consuming excessive computational resources during the encoding process.

[0040] The inference latency setting for LSTM autoencoder models is required to be less than 10 milliseconds. Inference latency refers to the time required for the model to complete calculations and output the reconstruction error after receiving input data. In vehicle-mounted warning scenarios, real-time performance is extremely critical, as changes in the braking system's state occur within a very short time. If the model's inference latency is too long, the warning signal output will be delayed, failing to promptly alert the driver to potential risks. Controlling the inference latency to within 10 milliseconds means the model can complete the processing of a pressure sequence window in a very short time. For example, if the duration of a sliding window is 0.5 seconds, the model can complete the reconstruction calculation for that window within 10 milliseconds, which is much shorter than the window's time interval. This ensures that the continuous processing of subsequent windows is not affected and that the warning signal is triggered immediately upon the occurrence of a risk, meeting the real-time operational requirements of vehicle-mounted systems.

[0041] The warning threshold is set as follows: Calculate the mean of the reconstruction error of normal training samples plus three times the standard deviation.

[0042] The LSTM autoencoder model can be replaced by a gated cyclic unit (GRU) or a Transformer encoder.

[0043] First, there's the replacement of LSTM with Gated Recurrent Units (GRUs). GRU is a simplified recurrent neural network structure that merges the forget gate and input gate of LSTM into an update gate, while simplifying the way unit states are passed. This reduces the number of parameters and computational complexity of the model while retaining the ability to capture temporal features. For example, in scenarios with limited computing power in automotive hardware, using GRU instead of LSTM can both encode and reconstruct braking pressure time-series data, further reducing the model's inference latency. This allows the warning system to run stably on low-computing-power devices, while ensuring that the learning effect on pressure change patterns is not significantly reduced.

[0044] The Transformer encoder relies on a self-attention mechanism, enabling it to simultaneously capture the correlations between data at different locations within a sequence. Compared to the LSTM's sequential processing over time steps, it excels at capturing global features in long sequences. For instance, when pressure changes in a braking system exhibit correlations spanning extended periods, the Transformer encoder can directly locate these distant correlations through self-attention weights, resulting in more accurate calculations of reconstruction errors.

[0045] The basic units of an LSTM autoencoder model include forget gate, input gate, and output gate structures; The activation control of the forget gate, input gate, and output gate is determined by... Function implementation; The input gate also includes a function for generating candidate cell states. layer; At each time step, the hidden state of the cell in the previous time step. Unit status and current input For input; The forget gate controls the state of historical units. The degree of retention is controlled by the input gate for the current candidate state. The update degree of ' is used to calculate the current cell state. ; The output gate depends on the current cell state. and hidden state Control the output of the current time step .

[0046] The basic LSTM unit comprises three structures: a forget gate, an input gate, and an output gate. The activation control of these three gates is implemented using the Sigmoid function. The output of the Sigmoid function ranges from 0 to 1, where 0 represents completely closing the corresponding gate and 1 represents completely opening it, thus flexibly controlling the degree of information transmission. The input gate is configured with... layer, The layer's output ranges from -1 to 1 and is used to generate candidate cell states.

[0047] At each time step, the cell takes the hidden state and cell state of the previous time step, as well as the current input pressure data, as input. Assuming the current input is the pressure value of a certain sampling point, the hidden state of the previous time step stores the feature information of the previous pressure data, and the cell state is the accumulated historical information.

[0048] The forget gate controls the degree to which historical cell states are retained. For example, when the braking process transitions from the pressure rise phase to the steady phase, the forget gate outputs a value close to 1, retaining the characteristic information of the pressure rise; while when the braking ends and the pressure drops rapidly, the forget gate outputs a value close to 0, weakening the information from the previous steady phase and retaining only the content related to the current downward trend.

[0049] The input gate controls the degree of update to the current candidate cell state. The input gate... The layer generates new candidate cell states based on the current input and the hidden state of the previous time step. The sigmoid output of the input gate determines the proportion of candidate states that are updated into the current cell state. For example, when the pressure value of the current input fluctuates slightly, the input gate outputs a smaller value, allowing a small number of candidate states to be updated into the cell state, thus avoiding random fluctuations from interfering with the overall information.

[0050] By using a forget gate to filter historical states and an input gate to select candidate states, the cell computation obtains the cell state for the current time step. The output gate controls the output for the current time step based on the current cell state and the hidden state. The sigmoid output of the output gate performs a tanh transformation on the current cell state, and then multiplies it with its own output to obtain the hidden state for the current time step. This hidden state serves as one of the inputs for the next time step and also participates in subsequent sequence reconstruction calculations.

[0051] As the pressure slowly increases from its initial value, the LSTM unit retains the baseline information of the initial pressure through a forget gate, introduces the pressure change characteristics of the rising phase through the input gate, and then passes the integrated features to the next time step through the output gate. When the pressure enters the stable phase, the forget gate retains the features of the rising phase, the input gate introduces the stable features of the stable phase, and the output gate passes the information of the stable state. It is through this cyclical transmission that the LSTM unit can completely capture the full time-series features of braking pressure from rising to stabilizing and then falling, providing a reliable feature foundation for subsequent model reconstruction of the pressure sequence and calculation of reconstruction errors.

[0052] The following are also included before triggering the oil leak warning signal: Calculate the distance between the pressure data characteristics and the curves of different disturbance modes to determine whether the disturbance source is located inside the monitored pipe section; If the disturbance source is determined to be located within the pipe section, an oil leak warning signal will be triggered. If the disturbance source is determined to be outside the pipe section, it is identified as external equipment operation and will not trigger an oil leak warning signal.

[0053] By calculating the distance between the pressure data characteristics and the curves of different disturbance patterns, it can be determined whether the disturbance source is located inside the monitored pipe section. The disturbance patterns are pre-prepared characteristic curves for various scenarios, such as the pressure drop curve when oil leaks inside the pipe section or the pressure fluctuation curve when external equipment is operated. Each pattern corresponds to a specific pressure change pattern. The curve distance refers to the degree of difference between the characteristic curve of the real-time pressure data and these preset disturbance pattern curves; the smaller the distance, the higher the matching degree between the real-time data and the corresponding disturbance pattern.

[0054] In practice, features of real-time pressure data are extracted, such as the rate of pressure change, fluctuation amplitude, and duration, to form a corresponding feature curve. The distance between this curve and the preset leakage disturbance mode curve within the pipe section and the external equipment operation disturbance mode curve are calculated respectively.

[0055] When a slight leak occurs in the braking system line while the vehicle is in motion, the pressure will show a slow and continuous downward trend, and its characteristic curve will be close to the preset leakage disturbance mode curve in the line. However, if the pressure fluctuation is caused by the driver accidentally touching the brake pedal, the pressure will rise rapidly and then recover quickly, and the corresponding characteristic curve will be closer to the disturbance mode curve of external equipment operation.

[0056] If the disturbance source is determined to be inside the pipe section, it indicates that the anomaly is caused by oil leakage from the braking system itself, triggering an oil leakage warning signal. If the disturbance source is determined not to be inside the pipe section, it indicates that the pressure fluctuation is a normal disturbance caused by external operation, not a system malfunction, and therefore no warning signal will be triggered. This determination process effectively distinguishes between internal leaks and external interference, avoiding false alarms caused by external operations and making the warning results more accurate and reliable.

[0057] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0058] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0059] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of fluid leakage in a vehicle braking system based on LSTM, characterized in that, The method includes the following steps: The pressure data of the master cylinder or wheel cylinder is collected by the pressure sensor of the vehicle braking system to form the raw pressure timing data; The original pressure time series data is denoised and normalized, and a sliding window is used to divide the processed original pressure time series data into continuous pressure sequence windows, each pressure sequence window containing N sampling points; An LSTM autoencoder model is constructed, using a stress sequence window under normal operating conditions as training data. The training aims to minimize the reconstruction error, and an early warning threshold is set based on the reconstruction error distribution of the training set. Real-time pressure data is collected, preprocessed, and then input into a trained LSTM autoencoder model to calculate the reconstruction error; an oil leakage early warning signal is triggered based on the reconstruction error. If the reconstruction error exceeds the threshold and continues for a preset number of windows, an oil leak warning signal will be triggered. The warning signal is triggered by outputting through the vehicle communication system.

2. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The oil leakage early warning signal is triggered based on the reconstruction error, specifically as follows: If the reconstruction error exceeds the threshold and continues for a preset number of windows, an oil leak warning signal will be triggered.

3. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The warning signal triggered by the vehicle communication system includes the following steps: The warning signal is output through at least one of the following methods: CAN bus, vehicle terminal, instrument light, buzzer, or vehicle networking platform.

4. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The pressure sensor has a sampling frequency of 100Hz; The length of the sliding window is set to 50 sampling points, corresponding to a duration of 0.5 seconds.

5. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The original pressure time series data were denoised using moving average or wavelet filtering. The raw pressure time series data were normalized using Z-score or Min-Max methods.

6. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The hidden layer dimension of the LSTM autoencoder model is ≤64, and the inference latency of the LSTM autoencoder model is <10ms.

7. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The warning threshold is set as follows: Calculate the mean of the reconstruction error of normal training samples plus three times the standard deviation.

8. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 1, characterized in that, The LSTM autoencoder model can be replaced by a gated cyclic unit (GRU) or a Transformer encoder.

9. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 8, characterized in that, The basic units of an LSTM autoencoder model include forget gate, input gate, and output gate structures; The activation control of the forget gate, input gate, and output gate is determined by... Function implementation; The input gate also includes a function for generating candidate cell states. layer; At each time step, the hidden state of the cell in the previous time step. Unit status and current input For input; The forget gate controls the state of historical units. The degree of retention is controlled by the input gate for the current candidate state. The update degree of ' is used to calculate the current cell state. ; The output gate depends on the current cell state. and hidden state Control the output of the current time step .

10. The method for early warning of fluid leakage in a vehicle braking system based on LSTM according to claim 2, characterized in that, The following are also included before triggering the oil leak warning signal: Calculate the distance between the pressure data characteristics and the curves of different disturbance modes to determine whether the disturbance source is located inside the monitored pipe section; If the disturbance source is determined to be located within the pipe section, an oil leak warning signal will be triggered. If the disturbance source is determined to be outside the pipe section, it is identified as external equipment operation and will not trigger an oil leak warning signal.

Images