Intelligent engine oil state monitoring method and system based on multi-parameter sensing and deep learning

Abstract

The invention belongs to the field of engine oil state monitoring, and discloses an intelligent engine oil state monitoring method and system based on multi-parameter sensing and deep learning. Preprocessing the collected engine oil physical and chemical data; inputting the engine oil physical and chemical data into the bidirectional LSTM engine oil life prediction model, and dynamically generating an engine oil remaining use cycle evaluation value through the bidirectional LSTM engine oil life prediction model; and judging the oil level of the residual use cycle evaluation check-in machine of the engine oil. According to the intelligent engine oil state monitoring method and system integrating multi-parameter real-time sensing, dynamic data compensation, edge calculation optimization, cloud intelligent analysis and graded early warning linkage, the limitation that a traditional engine oil detection means is single in function, lags behind time and depends on experience judgment is broken through; a conventional engine oil monitoring device is upgraded into an intelligent operation and maintenance platform integrating state sensing, service life prediction, fault early warning and maintenance decision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of engine oil condition monitoring, in particular to an intelligent engine oil condition monitoring method and system based on multi-parameter sensing and deep learning. BACKGROUND

[0002] Traditional engine oil condition monitoring methods rely on periodic manual detection or single parameter threshold alarms, which have obvious monitoring lags, lack comprehensive performance evaluations, and have insufficient life prediction accuracy, making it difficult to meet the fine maintenance needs of modern equipment under high load, long cycle, and multiple working conditions. Current equipment maintenance generally faces problems such as unclear engine oil degradation mechanism, lack of sudden failure warning, and reliance on experience for maintenance cycle. How to real-time perceive the evolution of engine oil physical and chemical properties, accurately predict the remaining useful life, and dynamically develop scientific maintenance strategies to avoid equipment abnormal wear, energy consumption surge, and even major safety accidents caused by engine oil failure has become a core issue that needs to be addressed in the intelligent operation and maintenance system of industrial equipment. SUMMARY

[0003] The present application aims to provide an intelligent engine oil condition monitoring method and system based on multi-parameter sensing and deep learning to solve the problems mentioned in the background.

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] An intelligent engine oil condition monitoring method based on multi-parameter sensing and deep learning, comprising:

[0006] Step 1: Collecting engine oil physical and chemical data, including viscosity, moisture, temperature, and liquid level.

[0007] Step 2: Preprocessing the collected engine oil physical and chemical data.

[0008] Step 3: Inputting the preprocessed engine oil physical and chemical data into a bidirectional LSTM engine oil life prediction model to dynamically generate an engine oil remaining useful life evaluation value.

[0009] Step 4: Judging the engine oil remaining useful life evaluation value and engine oil level. When the engine oil remaining useful life evaluation value decays to a critical threshold or the engine oil level deviates significantly from the safety interval, a structured warning instruction is automatically generated and sent to the edge computing node, triggering the terminal device to execute the preset alarm protocol.

[0010] Further, the step 2 comprises:

[0011] Step 2.1: Correcting the system error caused by temperature through a dynamic temperature compensation formula wherein, μ comp is the compensated value, and μraw is the original measured value, e is a natural constant, K is a temperature coefficient, T is the current temperature, and T0 is a reference temperature;

[0012] Step 2.2, using Kalman filtering, based on the state equation predicts the current state, where, is the prior estimate of the state at time k given the information at time k-1, and F is the state transition matrix, is the posterior state estimate at time k-1; and the optimal estimate is updated by the observation z k comp and the Kalman gain K k , P k|k-1 is the covariance of the prior estimation error given the information at time k-1, and R represents the observation noise. Finally, a high-precision and low-lag viscosity value is output.

[0013] Further, the working process of the oil life prediction model in step 3 includes:

[0014] The preprocessed oil physicochemical data is input into the remaining service period evaluation value oil level for life prediction. The oil life prediction uses a bidirectional LSTM network to extract time sequence features, and the hidden state update formula is:

[0015]

[0016] is the hidden state of the forward LSTM at time steps t and t-1, LSTM is the internal calculation function of the LSTM unit, x t is the input feature vector at time step t, and T is the total number of time steps;

[0017] The time step weight is calculated through the attention mechanism:

[0018]

[0019] u t is the intermediate value of the attention score at time step t, tanh is the activation function, W a is the full connection layer weight matrix of the attention mechanism, b a is the full connection layer bias vector of the attention mechanism, alpha t is the normalized attention weight at time step t, exp is the natural exponential function, u i is the attention score at time step i summed over the index i;

[0020] The context vector is generated:

[0021] ​​

[0022] c is a context vector, The hidden state of the backward LSTM at time step t;

[0023] The final output of the Sigmoid activation function is the remaining life percentage:

[0024]

[0025] L life is the final prediction output value of the bidirectional LSTM engine life prediction model, σ is the Sigmoid activation function, W fc is the weight matrix of the output layer, b f is the bias term of the output layer, σ(z) represents performing the Sigmoid function operation on the variable z, z is the output value of the linear layer, e is the natural constant;

[0026] The weighted cross-entropy loss function is used for training the bidirectional LSTM engine life prediction model:

[0027]

[0028] τ is the loss value, N is the total number of samples, y i is the true remaining life percentage of the i-th sample, log is the natural logarithm, is the remaining life percentage predicted by the model for the i-th sample.

[0029] The training process is optimized by combining the AdamW optimizer, Dropout, and early stopping mechanism;

[0030] When deployed online, the bidirectional LSTM engine life prediction model is quantized to INT8 precision by the TensorRT engine, and supports incremental learning updates. The parameter expression used to update the model during training is:

[0031]

[0032] W new is the weight of the updated bidirectional LSTM engine life prediction model, W old is the weight of the bidirectional LSTM engine life prediction model before updating, η inc is the learning rate, is the gradient of the loss function τ with respect to the weight W, X new is the new input data batch.

[0033] Further, it further includes step 5, comprising:

[0034] When the engine life prediction accuracy is insufficient, manual review is triggered, and when the remaining life is lower than the preset minimum life threshold, the replacement instruction is forced to push.

[0035] The application also includes an intelligent machine oil state monitoring system based on multi-parameter sensing and deep learning, which is used to realize the intelligent machine oil state monitoring method as described above, and includes a data acquisition module, a communication module, a data processing module, a cloud server, a power module, and a display.

[0036] The data acquisition module acquires the physicochemical data of the machine oil in real time.

[0037] The main control module receives the physicochemical data of the machine oil collected by the data acquisition module and processes the physicochemical data of the machine oil.

[0038] The communication module transmits the processed physicochemical data of the machine oil from the main control module to the cloud server.

[0039] After receiving the uploaded physicochemical data of the machine oil from the main control module, the cloud server performs the remaining service life evaluation of the lubricating oil through the deployed bidirectional LSTM lubricating oil life prediction model, dynamically generates the remaining service life evaluation value based on the multi-dimensional data fusion bidirectional LSTM lubricating oil life prediction model, and when detecting that the lubricating oil life decays to the critical threshold or the liquid level parameter deviates from the safety interval, automatically generates a structured early warning instruction and issues it to the edge computing node to trigger the terminal device to execute the preset alarm protocol.

[0040] The display shows the remaining life percentage and the current liquid level state of the machine oil in real time.

[0041] Further, the data acquisition module includes a viscosity sensor, a moisture sensor, a temperature sensor, and a liquid level sensor.

[0042] Further, the power module charges the battery of the vehicle when the vehicle is in the starting or power supply state.

[0043] Compared with the prior art, the beneficial effects of the present application are: the present application aims to break through the limitations of single function, time lag and experience-dependent judgment of traditional engine oil detection methods by an intelligent engine oil state monitoring method and system integrating multi-parameter real-time perception, dynamic data compensation, edge computing optimization, cloud intelligent analysis and hierarchical early warning linkage, so that the conventional engine oil monitoring device is upgraded to an intelligent operation and maintenance platform integrating state perception, life prediction, fault warning and maintenance decision-making. The present application analyzes the physicochemical performance degradation law of engine oil through multi-source sensing data fusion, constructs a life prediction model based on deep learning to realize accurate quantization of maintenance period, eliminates environmental interference by combining with a working condition adaptive compensation algorithm, realizes "end-cloud-end" full-link closed-loop management relying on a bidirectional communication architecture, significantly improves the health management level of the lubricating system of the engine and industrial equipment, effectively prevents abnormal wear, energy efficiency decline and sudden failure caused by engine oil degradation, thereby prolongs the service life of the equipment, reduces the maintenance cost, avoids the risk of unplanned downtime, and provides intelligent protection for safe and economic operation of the equipment. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 It is a flow chart of an intelligent engine oil state monitoring method of multi-parameter sensing and deep learning.

[0045] Figure 2 It is a principle diagram of an intelligent engine oil state monitoring method of multi-parameter sensing and deep learning.

[0046] Figure 3 It is a display diagram of the remaining life percentage and the current liquid level state of the display.

[0047] Figure 4 It is a structural schematic diagram of an intelligent engine oil state monitoring system of multi-parameter sensing and deep learning.

[0048] Figure 5 It is a principle diagram of an intelligent engine oil state monitoring method. DETAILED DESCRIPTION

[0049] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0050] Please refer to Figures 1-3 FIG. 1 is an intelligent engine oil state monitoring method of multi-parameter sensing and deep learning, which comprises:

[0051] Step 1, collecting engine physicochemical data, the collected engine physicochemical data including viscosity, moisture, temperature and liquid level;

[0052] Step 2, preprocessing the collected engine oil physicochemical data, including:

[0053] Step 2.1, correcting the system error caused by temperature through a dynamic temperature compensation formula where μ comp is the compensated value μ raw is the original measured value, e is the natural constant, K is the temperature coefficient, T is the current temperature, and T0 is the reference temperature; Step 2.2, using Kalman filtering based on the state equation to predict the current state, where is the prior estimate of the state at time k under the information at time k-1, F is the state transition matrix, is the posterior state estimate at time k-1. And through the observation value z k = μ comp update the optimal estimate and Kalman gain K k , P k|k-1 covariance of the prior estimate error under the information at time k-1, R represents the observation noise, and finally output the viscosity value with high precision and low lag.

[0054] Step 3, input the preprocessed engine oil physicochemical data into the bidirectional LSTM engine oil life prediction model to dynamically generate the engine oil remaining use cycle evaluation value through the bidirectional LSTM engine oil life prediction model.

[0055] Wherein, the working process of the engine oil life prediction model includes:

[0056] Input the preprocessed engine oil physicochemical data into the bidirectional LSTM engine oil life prediction model for life prediction, the bidirectional LSTM engine oil life prediction model uses a bidirectional LSTM network to extract time series features, and the hidden state update formula is:

[0057]

[0058] is the hidden state of the forward LSTM at time steps t and t-1, LSTM is the internal calculation function of the LSTM unit, x t is the input feature vector at time step t.

[0059] And calculate the time step weight through the attention mechanism:

[0060]

[0061] u t is the attention score intermediate value at time step t, tanh is the activation function, Wa is the weight matrix of the fully connected layer of the attention mechanism, b a is the bias vector of the fully connected layer of the attention mechanism. α t is the normalized attention weight at time step t, exp is the natural exponential function, u i is the attention score at time step i summed over the index.

[0062] Generate the context vector:

[0063]

[0064] c is the context vector, is the hidden state of the backward LSTM at time step t.

[0065] Finally, the remaining life percentage is output through the Sigmoid activation function:

[0066]

[0067] L life is the final predicted output value of the model, σ is the Sigmoid activation function, W fc is the weight matrix of the output layer, b f is the bias term of the output layer. σ(z) represents performing the Sigmoid function operation on the variable z, z is the output value of the linear layer, e is the natural constant.

[0068] The model training uses a weighted cross-entropy loss function:

[0069]

[0070] τ is the loss value, N is N samples, y i is the true remaining life percentage of the i-th sample, log is the natural logarithm, is the remaining life percentage predicted by the model for the i-th sample.

[0071] The training process is optimized by combining the AdamW optimizer, Dropout, and early stopping mechanism;

[0072] When deployed online, the oil life prediction model is quantized to INT8 precision by the TensorRT engine, and supports incremental learning updates. The parameter expression used to update the model in training is:

[0073]

[0074] W new is the weight of the updated bidirectional LSTM oil life prediction model, W old is the weight of the bidirectional LSTM oil life prediction model before updating, ηinc is a learning rate, is a gradient of the loss function τ with respect to the weights W, X new is a new batch of input data.

[0075] Step 4, judging the engine oil level of the remaining service life evaluation value, when detecting that the remaining service life evaluation value decays to a critical threshold or the height of the engine oil level deviates from the safety interval, automatically generating a structured early warning instruction and issuing it to the edge computing node, triggering the terminal device to execute the preset alarm protocol.

[0076] Step 5, including: triggering manual review when the engine oil life prediction accuracy is insufficient, and forcibly pushing a replacement instruction when the remaining life is lower than the preset minimum life threshold.

[0077] As shown in Figure 4 and Figure 5 , an intelligent engine oil state monitoring system based on multi-parameter sensing and deep learning includes an oil level gauge module, a viscosity sensor, a moisture sensor, a temperature sensor, a liquid level sensor, an STM32 master control module, a SIM7600 communication module, a cloud server, a display module, and a power module. The power module input is connected to the vehicle power supply, and the output is connected to the oil level gauge module, the viscosity sensor, the moisture sensor, the temperature sensor, the liquid level sensor, the STM32 master control module, the SIM7600 communication module, the cloud server, and the display module. In addition to the oil level gauge module, the viscosity sensor, the moisture sensor, the temperature sensor, and the liquid level sensor being installed in the engine oil level gauge pipe, the remaining modules are installed in the cab.

[0078] In this system, the STM32 master control module selects an STM32F4 series high-performance microcontroller. This chip is based on the ARM Cortex-M4 core, has a main frequency of 168MHz and integrates a hardware floating point operation unit (FPU), has excellent real-time computing power and low power consumption characteristics, and can efficiently process multi-sensor data fusion and complex algorithm operations. The master control module is equipped with a FreeRTOS real-time operating system, which splits data acquisition (viscosity, moisture, temperature, liquid level), dynamic temperature compensation filtering, encrypted communication, and display control functions into independent tasks, realizes multi-task parallel processing through priority scheduling mechanism, and can timely perceive the changes of engine oil state.

[0079] The various sensors of the multi-parameter sensing and intelligent engine oil condition monitoring system based on deep learning are designed in a modular manner. The viscosity sensor adopts a JCV-2000 module, with a detection range of 600 kg / m3 to 1300 kg / m3. The moisture sensor is based on a JWA2 module, which detects water content. The digital temperature sensor uses a DS18B20 module, with a detection range of -40 degrees Celsius to 150 degrees Celsius. The liquid level sensor is equipped with a QDY30A-B-G-SGN module, with a detection range of 0 to 300 millimeters. The above modules can be quickly plugged in and out through standardized interfaces. The modular design significantly simplifies fault diagnosis and maintenance processes, and fully covers the real-time monitoring needs of core parameters such as engine oil viscosity, moisture, temperature, and liquid level.

[0080] The STM32 master module receives real-time multi-dimensional data streams from the viscosity sensor, moisture sensor, digital sensor, and liquid level sensor. Through dynamic temperature compensation algorithms and Kalman filter technology, the original data is denoised and calibrated. The edge computing unit performs local feature extraction and anomaly detection, and the preprocessed time series data is packaged and transmitted to the SIM76004G communication module through an encryption protocol. The processed engine oil viscosity, moisture, temperature, and liquid level data are uploaded to the cloud server.

[0081] The power module integrates dynamic voltage monitoring function. Through the built-in multi-stage voltage regulation unit, it implements dynamic voltage stabilization processing on the vehicle power input, effectively suppressing the impact of vehicle power grid voltage fluctuation on sensitive electronic components. The system adopts dual-mode intelligent power supply strategy: when detecting valid vehicle power input, it automatically switches to priority power mode, and realizes intelligent charging of energy storage unit; in the case of abnormal or disconnected vehicle power, it seamlessly switches to backup battery power mode, ensuring continuous and stable operation of the system.

[0082] The SIM7600 communication module is a full-band 4G LTE wireless communication module that supports major global operator networks and has full-duplex data transmission capability, compatible with multiple industrial communication protocols. The module adopts industrial-level packaging design, passes international certification standards, and integrates dual-antenna interface to enhance signal stability, suitable for extreme environments such as high temperature and vibration. It has built-in intelligent network switching and breakpoint resume functions, automatically switches to low-power mode in weak signal conditions, ensuring real-time data upload to the cloud and receiving maintenance instructions simultaneously, providing all-weather remote monitoring and instant decision support for equipment.

[0083] The cloud server receives the multi-dimensional sensor data set uploaded by the STM32 microcontroller, performs quantitative analysis of the degree of lubricating oil deterioration through a deployed machine learning prediction engine, dynamically generates a remaining use cycle evaluation value based on a multi-dimensional data fusion life prediction model, and automatically generates a structured early warning instruction and sends it to the edge computing node when detecting that the lubricating medium life has decayed to a critical threshold or the liquid level parameter deviates from the safety interval, triggering the terminal device to execute the preset alarm protocol.

[0084] The display module, as Figure 5 The display module receives and visually presents the lubricating oil remaining life quantitative indicators and oil storage dynamic parameters forwarded by the microcontroller. When the cloud early warning instruction reaches the control unit, it synchronously triggers the multi-modal interactive alarm mechanism, highlights the maintenance type code through the graphical interface, and prompts the driver to perform the corresponding maintenance operation with a beeping warning sound.

[0085] Embodiments:

[0086] The oil level gauge module and various sensors are installed in the oil level gauge pipe of the vehicle to detect the state of the oil, and are connected to the power module. The oil level gauge module collects real-time oil variables and sends them to the main control module for data preprocessing, and then sends the preprocessed data to the cloud server

[0087] The cloud server predicts the oil life based on the uploaded sensor information and judges in real time whether the liquid level reaches the warning threshold. If the threshold is not reached, the vehicle system returns the oil life percentage and liquid level percentage data. Once the oil life or liquid level reaches the warning threshold, the server not only returns the percentage data of the two, but also sends the corresponding oil life warning information or oil liquid level warning information, and triggers a sound alarm to remind the driver to replace the oil or supplement the oil in time.

[0088] Although embodiments of the present application have been shown and described, it is to be understood that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent oil condition monitoring using multi-parameter sensing and deep learning, characterized in that, include: Step 1: Collect engine oil physicochemical data, including viscosity, moisture content, temperature, and level. Step 2: Preprocess the collected physical and chemical data of the engine oil; Step 3: Input the preprocessed physicochemical data of the engine oil into the bidirectional LSTM engine oil life prediction model, and dynamically generate the remaining service life assessment value of the engine oil through the bidirectional LSTM engine oil life prediction model. Step 4: Determine the oil level based on the remaining service life assessment value. When the remaining service life assessment value is detected to have decayed to a critical threshold or the oil level deviates significantly from the safe range, a structured warning command is automatically generated and sent to the edge computing node, triggering the terminal device to execute the preset alarm protocol.

2. The intelligent oil condition monitoring method based on multi-parameter sensing and deep learning according to claim 1, characterized in that, Step 2 includes: Step 2.1, using the dynamic temperature compensation formula Correcting systematic errors caused by temperature, where μ comp This is the compensated value, μ raw These are the original measured values, e is the natural constant, K is the temperature coefficient, T is the current temperature, and T0 is the reference temperature; Step 2.2: Kalman filtering is used, based on the state equation. Predict the current state, where, Given the information at time k-1, F is the prior estimate of the state at time k, where F is the state transition matrix. It is the posterior state estimate at time k-1; and through the observed value z k =μ comp Update the optimal estimate and Kalman gain K k , P k|k-1 Given the information at time k-1, the covariance of the prior estimation error, where R represents the observation noise, ultimately outputs a high-precision, low-hysteresis viscosity value.

3. The intelligent oil condition monitoring method based on multi-parameter sensing and deep learning according to claim 1, characterized in that, The working process of the oil life prediction model in step 3 includes: The preprocessed physicochemical data of the engine oil are input into the remaining service life assessment value and engine oil level for life prediction. The engine oil life prediction uses a bidirectional LSTM network to extract temporal features, and the hidden state update formula is as follows: This represents the hidden state of the forward LSTM at time steps t and t-1. LSTM is the internal computation function of the LSTM unit. t The input feature vector at time step t, where T is the total number of time steps; And calculate the time step weights using an attention mechanism: u t is the median attention score at time step t, tanh is the activation function, and W a This is the weight matrix of the fully connected layer in the attention mechanism, b a It is the bias vector of the fully connected layer in the attention mechanism, α t The normalized attention weights are at time step t, where exp is the natural exponential function, and u i It is the attention score at the time step with summation index i; Generate context vectors: c is the context vector. The hidden state of the backward LSTM at time step t; Finally, the Sigmoid activation function outputs the remaining lifetime percentage: L life This is the final predicted output value of the bidirectional LSTM oil life prediction model, where σ is the Sigmoid activation function, and W... fc It is the weight matrix of the output layer, b f The bias term of the output layer, σ(z), represents the operation of the Sigmoid function on the variable z, where z is the output value of the linear layer, and e is the natural constant; The bidirectional LSTM oil life prediction model is trained using a weighted cross-entropy loss function: τ is the loss value, N is the total number of samples, and y i The true remaining lifespan percentage for the i-th sample is given by log, where log is the natural logarithm. The percentage of remaining lifespan predicted by the model for the i-th sample. The training process is optimized by combining the AdamW optimizer, Dropout, and early stopping mechanism. When deployed online, the bidirectional LSTM oil life prediction model is quantized to INT8 accuracy using the TensorRT engine and incremental learning updates are supported. The parameter expression used to update the model during training is as follows: W new The updated weights, W, of the bidirectional LSTM oil life prediction model. old The weights, η, of the previous bidirectional LSTM oil life prediction model. inc It's the learning rate. X is the gradient of the loss function τ with respect to the weights W. new This is a new batch of input data.

4. The intelligent oil condition monitoring method based on multi-parameter sensing and deep learning according to claim 1, characterized in that, It also includes step 5, which includes: Manual re-inspection is triggered when the oil life prediction accuracy is insufficient, and a replacement command is forcibly pushed when the remaining life is lower than the preset minimum life threshold.

5. A multi-parameter sensing and deep learning intelligent oil condition monitoring system, used to implement the intelligent oil condition monitoring method as described in any one of claims 1-4, characterized in that, It includes a data acquisition module, a communication module, a data processing module, a cloud server, a power supply module, and a display. The data acquisition module collects engine oil physicochemical data in real time; The main control module receives the physical and chemical data of the engine oil collected by the data acquisition module and processes the physical and chemical data of the engine oil. The communication module transmits the processed oil physicochemical data from the main control module to the cloud server. After receiving the physicochemical data of the engine oil uploaded by the main control module, the cloud server performs an assessment of the remaining service life of the lubricating oil through the deployed bidirectional LSTM engine oil life prediction model. The bidirectional LSTM engine oil life prediction model based on multi-dimensional data fusion dynamically generates the remaining service life assessment value. When the lubricating oil life decays to the critical threshold or the liquid level parameter deviates from the safe range, a structured early warning command is automatically generated and sent to the edge computing node, triggering the terminal device to execute the preset alarm protocol. The display shows the remaining lifespan percentage and current level of the engine oil in real time.

6. The intelligent oil condition monitoring system based on multi-parameter sensing and deep learning according to claim 5, characterized in that, The data acquisition module includes a viscosity sensor, a moisture sensor, a temperature sensor, and a liquid level sensor.

7. The intelligent oil condition monitoring system based on multi-parameter sensing and deep learning according to claim 5, characterized in that, It also includes a power module that charges its own battery when the vehicle is in a starting or powered state.

Images