Oil level and pressure monitoring and alarming integrated state monitoring method, system, equipment and storage medium

Abstract

The invention discloses a state monitoring method, system, equipment and storage medium integrating oil level and pressure monitoring and alarming, and relates to the technical field of circuits, oil level and pressure data are obtained through deep learning target detection and data analysis, and the method comprises the steps of accurately positioning and analyzing key features of images of an oil level gauge and a pressure gauge; during oil level monitoring, an oil level value is calculated through the steps of marking, recognizing, fitting and the like, and during pressure monitoring, the circle center of a dial plate, the position of a pointer and the like are marked and recognized to be converted into a pressure value; dynamically calibrating and adjusting an image recognition threshold value and a visual detection model by utilizing environment perception; a trend prediction model is embedded to realize intelligent alarm and prediction and early warning which are divided into first-level early warning and second-level early warning; and data storage and visual analysis provide a basis for operation and maintenance management. The system is composed of a sensing layer, a data processing layer, an analysis decision-making layer and an application layer. Compared with the prior art, deep learning and intelligent calculation are combined to improve the data analysis precision, and the operation and maintenance intelligent level is improved.

Description

Technical Field

[0001] The present invention relates to the field of circuit technology, and in particular to a state monitoring method, system, equipment, and storage medium integrating oil level and pressure monitoring and alarming. Background Art

[0002] Accurate monitoring of oil levels and pressure is crucial in many areas, including industrial production and equipment operation and maintenance. For example, in automotive engine systems, the proper oil level directly impacts the engine's lubrication and service life. Pressure monitoring can reflect the system's operating status and prevent malfunctions. In hydraulic systems, stable oil levels and pressure are key factors in ensuring proper equipment operation and maintaining production efficiency. With the acceleration of industrial automation, the need for real-time, accurate monitoring of equipment operating status is becoming increasingly urgent, and oil level and pressure monitoring technologies are constantly evolving.

[0003] Currently, oil level and pressure monitoring for substation equipment primarily relies on manual inspections or single sensor detection. These issues include large measurement errors, poor real-time performance, and limited remote management capabilities, making them difficult to meet the automated O&M requirements of unmanned substations. Furthermore, traditional monitoring systems lack intelligent recognition and precise calculation capabilities, and are easily affected by environmental factors such as light, temperature, and humidity. This results in unstable recognition and a high false alarm rate, impacting equipment operational safety and O&M efficiency.

[0004] To address the drawbacks of manual monitoring, monitoring methods based on sensor technology have emerged. Mechanical and electronic sensors can convert oil level and pressure signals into electrical or other measurable signals, enabling a degree of automated monitoring. However, these sensors have poor adaptability in complex environments. High temperatures, high humidity, or strong electromagnetic interference can affect sensor performance, leading to reduced measurement accuracy and even failure. Furthermore, traditional sensors often only monitor a single parameter and are unable to perform comprehensive analysis of oil level and pressure, making them unable to meet the modern industrial demand for comprehensive equipment monitoring.

[0005] With the rapid development of information technology, the Internet of Things (IoT), big data, and artificial intelligence (AI) are increasingly being applied to oil level and pressure monitoring. IoT technology enables remote data transmission and sharing, allowing managers to access equipment status information anytime, anywhere. Big data technology can store and analyze large amounts of monitoring data, unlocking its potential value. However, existing IoT and big data-based monitoring systems still face several challenges. Data accuracy and reliability need to be improved, with data noise and outliers impacting the credibility of analysis results. Furthermore, these systems lack intelligent early warning capabilities, with most systems only issuing alerts after equipment failures or parameter anomalies occur. This makes it difficult to predict potential risks and effectively mitigate losses caused by equipment failures.

[0006] Against this backdrop, the development of a condition monitoring technology that integrates oil level and pressure monitoring and alarms, offering high precision, adaptability, intelligent early warning, and remote management capabilities, is urgently needed. This technology must not only overcome the many shortcomings of traditional monitoring methods but also meet the demands of modern industrial intelligence and automation, providing strong support for stable equipment operation and efficient management. Summary of the Invention

[0007] The purpose of the present invention is to provide a condition monitoring method, system, equipment, and storage medium that integrates oil level and pressure monitoring and alarm. It obtains oil level and pressure data through deep learning target detection and data analysis, performs dynamic calibration through environmental perception, uses trend prediction models to achieve intelligent alarms and predictive warnings, and uses historical data storage for visual analysis to assist equipment operation and maintenance management to solve the above problems.

[0008] To achieve the above object, the present invention provides the following technical solutions:

[0009] A state monitoring method integrating oil level and pressure monitoring and alarming, characterized by comprising the following steps:

[0010] S1: Deep Learning Target Detection and Data Parsing: Using deep learning’s feature recognition capabilities, we accurately locate and analyze key features in oil level gauge and pressure gauge images, providing data support for subsequent measurement and analysis.

[0011] S2: Dynamic Calibration: Taking into account the complexity of actual application environments, by introducing an environmental perception mechanism, the image recognition threshold is dynamically adjusted and the visual detection model is optimized according to changes in environmental factors such as temperature, humidity, and lighting, so that the system can work stably and accurately in different environments.

[0012] S3: Intelligent Alarm and Predictive Warning: By embedding a trend prediction model to analyze oil level and pressure trends, the system upgrades from passive alarms to active warnings. This system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses.

[0013] S4: Data storage and visual analysis, stores historical data, and performs trend analysis and visual display. By analyzing historical data, the patterns and potential problems of equipment operation can be discovered, providing a decision-making basis for equipment maintenance and management. Visual display can present monitoring data and analysis results in an intuitive way, making it easier for users to view and understand.

[0014] Step S1, deep learning target detection and data parsing, specifically includes oil level monitoring and pressure monitoring. These two monitoring tasks are specific applications of deep learning target detection and data parsing to different monitoring objects. Feature recognition and data processing are performed on the oil level gauge and pressure gauge, respectively, to obtain relevant information about the oil level and pressure.

[0015] The oil level monitoring specifically includes the following steps:

[0016] S111: Input an oil level gauge image; obtain the oil level gauge image for analysis, which is the starting data for the entire oil level monitoring process.

[0017] S112: Labeling rules: labeling the oil level value, scale 0, scale 20, scale 40, scale 60, and scale 80 in the picture according to the rules;

[0018] S113: Identify the positions of the oil level frame and the scale frame through a deep learning model, and determine the coordinates of the center point of the oil level frame and the center point of the scale frame; specifically, define the coordinates of the center point of the oil level frame as (x oil ,y oil ), the coordinates of the center point of the scale frame are Where i∈(0,20,40,60,80); the purpose of labeling is to provide clear target information for subsequent deep learning model training and recognition, so that the model knows the objects to be detected and their corresponding labels.

[0019] S114: All scale frame center points are sorted vertically to obtain the scale order, and a polynomial fitting method is used to generate a scale baseline function f(x); by sorting the scale frame center points, the order relationship of the scale can be determined. The polynomial fitting method can generate a function to describe the scale baseline based on the distribution pattern of these points, providing a benchmark for subsequent calculation of the oil level value.

[0020] S115: Get the center point of the oil level frame (x oil ,y oil ), calculate the intersection point of its vertical projection on the scale reference line (x proj ,y proj ); the projection intersection is calculated by the scale reference line function f(x) (x proj ,y proj )=(x oil ,f(x oil )); Determine the corresponding position of the oil level on the scale reference line for subsequent calculation of the oil level value.

[0021] S116: Determine the scale range where the intersection point is located, and calculate the oil level value V using an interpolation formula. The interpolation formula is:

[0022]

[0023] Among them, lower_scale represents the oil level value of the lower scale frame of the scale range where the intersection is located, and upper_scale represents the oil level value of the upper scale frame of the scale range where the intersection is located. lower_scale Represents the y coordinate of the center point of the scale frame below the scale range where the intersection point is located, y upper_scale The y-coordinate of the center point of the scale frame above the intersection point. By determining the scale range where the intersection point is located and using the interpolation formula, the oil level value can be calculated more accurately based on the proportional relationship between the scales.

[0024] The pressure monitoring specifically includes the following steps:

[0025] S121: Input a pressure gauge image; obtain a pressure gauge image for pressure monitoring as basic data for subsequent analysis.

[0026] S122: Label the pointers and dials in the image according to the labeling rules; labeling the pointers and dials is to enable the deep learning model to recognize these key objects and prepare for subsequent determination of the pointer position and angle calculation.

[0027] S123: Identify the dial center and pointer position of the pressure gauge through deep learning model and Hough circle transform, and set the dial center (c x ,c y ), pointer center point (p x ,p y ); the deep learning model is responsible for identifying the approximate positions of the hands and dial, while the Hough circle transform can more accurately determine the position of the center of the dial. The combination of the two can accurately obtain the key position information of the hands and dial.

[0028] S124: Calculate the angle θ between the pointer and the horizontal right direction. The formula is:

[0029]

[0030] Where θ is the calculated angle in degrees, (c x ,c y ) is the coordinate of the dial center, (p x ,p y ) is the coordinate of the center point of the pointer; using the trigonometric function relationship, the angle between the pointer and the horizontal right direction is calculated through the coordinates of the pointer and the center of the dial, and the pointer position is converted into angle information.

[0031] S125: Correct the angle θ to the range [0,360]. If θ<0, correct it to θ=θ+360; ensure that the calculated angle is within a standard range.

[0032] S126: Set the effective angle range of the pressure gauge, map the angle θ to the pressure value, and use the normalization formula to convert the angle θ into the actual value P of the pressure gauge. The specific formula is:

[0033]

[0034] Where P is the pressure value calculated based on the pointer position, min_value is the minimum value of the dial, and max_value is the maximum value of the dial; θ min is the minimum value of the effective angle range of the pressure gauge, θ max The maximum value of the effective angle range of the pressure gauge. According to the design characteristics of the pressure gauge, the angle is mapped to the pressure value according to a certain proportional relationship, realizing the conversion from angle information to actual pressure value.

[0035] The oil level monitoring and pressure monitoring are both imported into RGB format images.

[0036] Step S2 dynamic calibration specifically involves introducing an environmental perception mechanism, combining temperature, humidity, light and other sensor data, dynamically adjusting the image recognition threshold, and optimizing the visual detection model. Specifically, it includes the following steps:

[0037] S21: Input data, including temperature, humidity, and light intensity values collected through environmental sensor data; collect environmental data as a basis for subsequent adjustments and optimizations, because environmental factors will affect image quality and detection results.

[0038] S22: Parameter definition, define image preprocessing parameters, including contrast adjustment coefficient C, which ranges from 0.5 to 2.0, brightness adjustment coefficient B, which ranges from -50 to 50; define deep learning model confidence threshold, including initial confidence threshold T0 = 0.5, dynamically adjusted confidence threshold T dynamic ; Clarify various parameters and their value ranges, which are used to ensure accuracy in subsequent image preprocessing and model adjustment.

[0039] S23: Data collection and normalization. Real-time environmental data is acquired through temperature and humidity sensors and light sensors. Temperature, humidity, and light intensity are normalized to the range [0, 1] to facilitate subsequent calculations. The normalization formula is as follows:

[0040]

[0041] Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters; normalization processing can unify environmental data of different ranges into a standard range and eliminate the influence of data dimension.

[0042] S24: Image preprocessing parameter adjustment: Dynamically adjust image preprocessing parameters according to the normalized environmental data, including the contrast adjustment coefficient C, whose formula is:

[0043] C = 1.0 + (0.5 × normalized light intensity) - (0.2 × normalized humidity)

[0044] Brightness adjustment coefficient B, its formula is

[0045] B = -50 + (100 × normalized light intensity)

[0046] Dynamically adjust the contrast and brightness of the image based on changes in environmental data. When the light intensity is high, the contrast is reduced to avoid overexposure, and when the humidity is high, the contrast is increased to reduce blur. When the light intensity is low, the brightness is increased to improve clarity, and when the light intensity is high, the brightness is reduced to avoid overexposure, so as to optimize image quality and improve detection accuracy.

[0047] S25: Dynamically update the confidence threshold of the deep learning model. The confidence threshold adjustment rule includes the following steps:

[0048] Set the weight W of environmental factors on image quality enν :

[0049] W env =0.4×normalized light intensity + 0.3×normalized humidity + 0.3×normalized temperature

[0050] Then calculate the dynamic confidence threshold T dynamic , the calculation formula is:

[0051] T dynamic =T0+(W env ×0.3)

[0052] Among them, W enν is the weight of the impact of environmental factors on image quality, T0 is the initial confidence threshold;

[0053] When the environmental conditions are poor, lower the confidence threshold to reduce the missed detection rate; when the environmental conditions are good, increase the confidence threshold T dynamic To reduce the false alarm rate; dynamically adjust the confidence threshold of the deep learning model according to the comprehensive influence weight of environmental factors, balance the missed detection rate and false alarm rate, so that the system can maintain good detection performance in different environments.

[0054] S26: Image preprocessing and model inference, image preprocessing is performed, and the original image is preprocessed according to the adjusted contrast coefficient C and brightness coefficient B, including the following steps:

[0055] To make contrast adjustments:

[0056] I contrast =C×(I original -128)+128

[0057] Among them, I original Represents the pixel value of the original image, C is the contrast adjustment coefficient, I contrast is the pixel value after contrast adjustment;

[0058] To make brightness adjustments:

[0059] I brightness =I contrast +B

[0060] Among them, I contrast The pixel value after contrast adjustment, B is the brightness adjustment coefficient, I brightness is the pixel value after adjusting the brightness;

[0061] The preprocessed images are used as input to the deep learning model;

[0062] Perform model inference:

[0063] Use the adjusted confidence threshold T dynamic Perform target detection and recognition;

[0064] If the confidence level of the test result is lower than T dynamic , then ignore the test result; otherwise, retain it and process it further.

[0065] S27: Establish a feedback mechanism, including false alarm rate monitoring and adaptive learning. The false alarm rate monitoring counts the false alarm rate in real time and analyzes it in combination with environmental data. If the false alarm rate is high, the preprocessing parameters and confidence threshold adjustment rules are further optimized; the adaptive learning associates historical environmental data with the detection results and trains a lightweight regression model to predict the optimal preprocessing parameters and confidence threshold.

[0066] Step S3: Intelligent alarm and prediction warning. By embedding a trend prediction model, it analyzes the changing trends of oil level and pressure, realizing the upgrade from passive alarm to active warning. The system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses. Specifically, it includes the following steps:

[0067] S31: Input data, including real-time monitoring data of oil level and pressure values, historical change records of oil level and pressure, lower limits of oil level and pressure, warning time window, set time point sequence T = {t1, t2, ..., tn}, and collect corresponding oil level or pressure value sequence X = {x1, x2, ..., xn};

[0068] S32: Determine the trend prediction model and use the LSTM long short-term memory network prediction model to determine the user-set oil level safety lower limit Lmin and pressure safety lower limit Pmin; dynamically adjust the dynamic warning threshold based on the volatility of historical data;

[0069] S33: Data preprocessing, removing outliers, using interpolation or mean to fill missing values, and normalization to normalize the oil level and pressure data to the [0, 1] interval to facilitate model training. The formula is:

[0070]

[0071] Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters;

[0072] S34: Trend prediction model training, using the LSTM model for time series prediction. The input features are the normalized oil level and pressure data series; the output targets are the predicted oil level and pressure values for a period of time in the future. During the training process, the LSTM model is trained by dividing the historical data into training and test sets.

[0073] S35: Trend prediction: input current data and the most recent time series data into the LSTM model for prediction, predicting the oil level and pressure values in the future.

[0074] S36: Safety threshold judgment: if the predicted value is lower than the safety lower limit set by the user, an early warning is triggered;

[0075] The alarm type is set as level one warning or level two warning according to the distance between the predicted value and the threshold. A level one warning is triggered when the predicted value is close to the safety threshold, and a level two warning is triggered when the predicted value exceeds the safety threshold.

[0076] A status monitoring system that integrates oil level and pressure monitoring and alarm is characterized by including a perception layer, a data processing layer, an analysis and decision-making layer, and an application layer. The perception layer collects images of the oil level gauge and pressure gauge, uses sensors to collect temperature, humidity, and light data, and transmits data with the help of a wireless communication module; the data processing layer uses a deep learning target detection module to identify key features, the intelligent computing module performs high-precision measurement, and the dynamic calibration module optimizes detection based on environmental data; the analysis and decision-making layer analyzes the change trend through a trend prediction model, and the threshold judgment module compares the predicted value with the threshold to trigger the corresponding operation; the application layer notifies the operation and maintenance personnel through the remote alarm module when the data is abnormal, the data storage module saves historical data, and the visualization analysis module displays the monitoring results.

[0077] A terminal device, characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

[0078] A computer-readable storage medium, characterized in that a computer program is stored in the storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

[0079] This state-monitoring technology, which integrates oil level and pressure monitoring and alarms, collects and transmits images of oil level and pressure gauges, as well as environmental data such as temperature, humidity, and lighting, through the perception layer. The data processing layer uses a deep learning target detection module to locate and analyze key image features. The intelligent computing module completes oil level and pressure measurements, and the dynamic calibration module optimizes detection based on environmental data. The analysis and decision-making layer uses a trend prediction model to analyze oil level and pressure trends, and the threshold judgment module compares predicted values with thresholds. If any data is abnormal, the application layer notifies operations and maintenance personnel via the remote alarm module. Meanwhile, the data storage module saves historical data, and the visual analysis module displays monitoring results, enabling comprehensive monitoring of equipment status, intelligent alarms, and predictive warnings.

[0080] Compared with the prior art, the present invention has the following beneficial effects:

[0081] The combination of deep learning target detection and intelligent computing enables the system to automatically identify the oil level gauge scale and pressure gauge pointer, eliminating manual reading. Through polynomial fitting and geometric calculations, data analysis accuracy is greatly improved, effectively avoiding the errors that may be caused by manual readings, ensuring accurate and reliable monitoring data and providing a precise basis for determining equipment operating status.

[0082] By incorporating an environmental perception mechanism, the system can dynamically adjust image recognition thresholds and optimize recognition algorithms in real time based on changes in environmental parameters such as temperature, humidity, and lighting. This enables the system to operate stably in diverse external environments, effectively reducing false alarms caused by environmental factors, significantly enhancing monitoring stability, and reducing maintenance costs.

[0083] Intelligent early warning and trend prediction capabilities enable the system to transition from passive alarms to active warnings. The embedded fault trend prediction model can proactively detect abnormal changes in oil levels or pressure, allowing operators ample time to take action and prevent unexpected failures. This not only improves equipment operation and maintenance efficiency but also reduces the risks of production interruptions and safety incidents caused by equipment failures, ensuring continuous and safe production.

[0084] With wireless data transmission capabilities and the ability to push abnormal information via SMS and WeChat mini-programs, remote monitoring and management are possible. In unmanned scenarios, maintenance personnel can obtain device operating status information anytime and anywhere and respond to abnormal situations promptly, greatly improving the timeliness and flexibility of maintenance, expanding the scope of equipment monitoring, and reducing manpower investment.

[0085] The system supports the storage, trend analysis, and visualization of historical data, providing rich data support for equipment management. Through in-depth analysis of historical data, managers can identify potential patterns and issues in equipment operation, thereby optimizing management strategies and achieving more scientific and accurate equipment management. The intuitive and clear visualization allows managers to quickly access key information, further enhancing the intelligent level of operation and maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0086] Figure 1 This is a flow chart of a state monitoring method that integrates oil level and pressure monitoring and alarming in the present invention;

[0087] Figure 2 This is an architecture diagram of a status monitoring system that integrates oil level and pressure monitoring and alarming in the present invention. Specific implementation methods

[0088] The technical solutions in the embodiments of the present invention will be fully described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0089] like Figure 1-2 As shown, a state monitoring method integrating oil level and pressure monitoring and alarm is characterized by comprising the following steps:

[0090] S1: Deep Learning Target Detection and Data Parsing: Using deep learning’s feature recognition capabilities, we accurately locate and analyze key features in oil level gauge and pressure gauge images, providing data support for subsequent measurement and analysis.

[0091] S2: Dynamic Calibration: Taking into account the complexity of actual application environments, by introducing an environmental perception mechanism, the image recognition threshold is dynamically adjusted and the visual detection model is optimized according to changes in environmental factors such as temperature, humidity, and lighting, so that the system can work stably and accurately in different environments.

[0092] S3: Intelligent Alarm and Predictive Warning: By embedding a trend prediction model to analyze oil level and pressure trends, the system upgrades from passive alarms to active warnings. This system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses.

[0093] S4: Data storage and visual analysis, stores historical data, and performs trend analysis and visual display. By analyzing historical data, the patterns and potential problems of equipment operation can be discovered, providing a decision-making basis for equipment maintenance and management. Visual display can present monitoring data and analysis results in an intuitive way, making it easier for users to view and understand.

[0094] Step S1 deep learning target detection and data analysis specifically includes oil level monitoring and pressure monitoring.

[0095] The oil level monitoring specifically includes the following steps:

[0096] S111: Input oil level gauge picture;

[0097] S112: Labeling rules: labeling the oil level value, scale 0, scale 20, scale 40, scale 60, and scale 80 in the picture according to the rules;

[0098] S113: Identify the positions of the oil level frame and the scale frame through a deep learning model, and determine the coordinates of the center point of the oil level frame and the center point of the scale frame; specifically, define the coordinates of the center point of the oil level frame as (x oil ,y oil ), the coordinates of the center point of the scale frame are where i∈(0,20,40,60,80);

[0099] S114: sorting all the center points of the scale frames vertically to obtain a scale order, and generating a scale baseline function f(x) using a polynomial fitting method;

[0100] S115: Get the center point of the oil level frame (x oil ,y oil ), calculate the intersection point of its vertical projection on the scale reference line (x proj ,y proj ); the projection intersection is calculated by the scale reference line function f(x) (x proj ,y proj )=(x oil ,f(x oil ));

[0101] S116: Determine the scale range where the intersection point is located, and calculate the oil level value V using an interpolation formula. The interpolation formula is:

[0102]

[0103] Among them, lower_scale represents the oil level value of the lower scale frame of the scale range where the intersection is located, and upper_scale represents the oil level value of the upper scale frame of the scale range where the intersection is located. lower_scaleRepresents the y coordinate of the center point of the scale frame below the scale range where the intersection point is located, y upper_scale Represents the y-coordinate of the center point of the upper scale box of the scale range where the intersection point is located.

[0104] The pressure monitoring specifically includes the following steps:

[0105] S121: Input pressure gauge picture;

[0106] S122: Label the pointer and dial in the picture according to the labeling rules;

[0107] S123: Identify the dial center and pointer position of the pressure gauge through deep learning model and Hough circle transform, and set the dial center (c x ,c y ), pointer center point (p x ,p y );

[0108] S124: Calculate the angle θ between the pointer and the horizontal right direction. The formula is:

[0109]

[0110] Where θ is the calculated angle in degrees, (c x ,c y ) is the coordinate of the dial center, (p x ,p y ) is the coordinate of the center point of the pointer;

[0111] S125: Correct the angle θ to the range [0, 360]. If θ<0, correct it to θ=θ+360.

[0112] S126: Set the effective angle range of the pressure gauge, map the angle θ to the pressure value, and use the normalization formula to convert the angle θ into the actual value P of the pressure gauge. The specific formula is:

[0113]

[0114] Where P is the pressure value calculated based on the pointer position, min_value is the minimum value of the dial, and max_value is the maximum value of the dial; θ min is the minimum value of the effective angle range of the pressure gauge, θ max It is the maximum value of the effective angle range of the pressure gauge.

[0115] The oil level monitoring and pressure monitoring are both imported into RGB format images.

[0116] Step S2 dynamic calibration specifically involves introducing an environmental perception mechanism, combining temperature, humidity, light and other sensor data, dynamically adjusting the image recognition threshold, and optimizing the visual detection model. Specifically, it includes the following steps:

[0117] S21: Input data, including temperature, humidity, and light intensity values collected through environmental sensor data;

[0118] S22: Parameter definition, define image preprocessing parameters, including contrast adjustment coefficient C, which ranges from 0.5 to 2.0, brightness adjustment coefficient B, which ranges from -50 to 50; define deep learning model confidence threshold, including initial confidence threshold T0 = 0.5, dynamically adjusted confidence threshold T dynamic ;

[0119] S23: Data collection and normalization: Real-time environmental data is acquired through temperature and humidity sensors and light sensors. Temperature, humidity, and light intensity are normalized to the range [0, 1] to facilitate subsequent calculations. The normalization formula is as follows:

[0120]

[0121] Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters;

[0122] S24: Image preprocessing parameter adjustment: Dynamically adjust image preprocessing parameters according to the normalized environmental data, including the contrast adjustment coefficient C, whose formula is:

[0123] C = 1.0 + (0.5 × normalized light intensity) - (0.2 × normalized humidity)

[0124] Brightness adjustment coefficient B, its formula is

[0125] B = -50 + (100 × normalized light intensity)

[0126] S25: Dynamically update the confidence threshold of the deep learning model. The confidence threshold adjustment rule includes the following steps:

[0127] Set the weight W of environmental factors on image quality enν :

[0128] W env =0.4×normalized light intensity + 0.3×normalized humidity + 0.3×normalized temperature

[0129] Then calculate the dynamic confidence threshold T dynamic , the calculation formula is:

[0130] T dynamic=T0+(W env ×0.3)

[0131] Among them, W enν is the weight of the impact of environmental factors on image quality, T0 is the initial confidence threshold;

[0132] When the environmental conditions are poor, lower the confidence threshold to reduce the missed detection rate; when the environmental conditions are good, increase the confidence threshold T dynamic To reduce false alarm rates;

[0133] S26: Image preprocessing and model inference, image preprocessing is performed, and the original image is preprocessed according to the adjusted contrast coefficient C and brightness coefficient B, including the following steps:

[0134] To make contrast adjustments:

[0135] I contrast =C×(I original -128)+128

[0136] Among them, I original Represents the pixel value of the original image, C is the contrast adjustment coefficient, I contrast is the pixel value after contrast adjustment;

[0137] To make brightness adjustments:

[0138] I brightness =I contrast +B

[0139] Among them, I contrast The pixel value after contrast adjustment, B is the brightness adjustment coefficient, I brightness is the pixel value after adjusting the brightness;

[0140] The preprocessed images are used as input to the deep learning model;

[0141] Perform model inference:

[0142] Use the adjusted confidence threshold T dynamic Perform target detection and recognition;

[0143] If the confidence level of the test result is lower than T dynamic , then ignore the test result; otherwise, retain it and process it further.

[0144] S27: Establish a feedback mechanism, including false alarm rate monitoring and adaptive learning. The false alarm rate monitoring counts the false alarm rate in real time and analyzes it in combination with environmental data. If the false alarm rate is high, the preprocessing parameters and confidence threshold adjustment rules are further optimized; the adaptive learning associates historical environmental data with the detection results and trains a lightweight regression model to predict the optimal preprocessing parameters and confidence threshold.

[0145] Step S3: Intelligent alarm and prediction warning. By embedding a trend prediction model, it analyzes the changing trends of oil level and pressure, realizing the upgrade from passive alarm to active warning. The system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses. Specifically, it includes the following steps:

[0146] S31: Input data, including real-time monitoring data of oil level and pressure values, historical change records of oil level and pressure, lower limits of oil level and pressure, warning time window, set time point sequence T = {t1, t2, ..., tn}, and collect corresponding oil level or pressure value sequence X = {x1, x2, ..., xn};

[0147] S32: Determine the trend prediction model and use the LSTM long short-term memory network prediction model to determine the user-set oil level safety lower limit Lmin and pressure safety lower limit Pmin; dynamically adjust the dynamic warning threshold based on the volatility of historical data;

[0148] S33: Data preprocessing, removing outliers, using interpolation or mean to fill missing values, and normalization to normalize the oil level and pressure data to the [0, 1] interval to facilitate model training. The formula is:

[0149]

[0150] Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters;

[0151] S34: Trend prediction model training, using the LSTM model for time series prediction. The input features are the normalized oil level and pressure data series; the output targets are the predicted oil level and pressure values for a period of time in the future. During the training process, the LSTM model is trained by dividing the historical data into training and test sets.

[0152] S35: Trend prediction: input current data and the most recent time series data into the LSTM model for prediction, predicting the oil level and pressure values in the future.

[0153] S36: Safety threshold judgment: if the predicted value is lower than the safety lower limit set by the user, an early warning is triggered;

[0154] The alarm type is set as level one warning or level two warning according to the distance between the predicted value and the threshold. A level one warning is triggered when the predicted value is close to the safety threshold, and a level two warning is triggered when the predicted value exceeds the safety threshold.

[0155] A status monitoring system that integrates oil level and pressure monitoring and alarm is characterized by including a perception layer, a data processing layer, an analysis and decision-making layer, and an application layer. The perception layer collects images of the oil level gauge and pressure gauge, uses sensors to collect temperature, humidity, and light data, and transmits data with the help of a wireless communication module; the data processing layer uses a deep learning target detection module to identify key features, the intelligent computing module performs high-precision measurement, and the dynamic calibration module optimizes detection based on environmental data; the analysis and decision-making layer analyzes the change trend through a trend prediction model, and the threshold judgment module compares the predicted value with the threshold to trigger the corresponding operation; the application layer notifies the operation and maintenance personnel through the remote alarm module when the data is abnormal, the data storage module saves historical data, and the visualization analysis module displays the monitoring results.

[0156] A terminal device, characterized in that it includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

[0157] A computer-readable storage medium, characterized in that a computer program is stored in the storage medium, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

[0158] The specific implementation method is as follows: S1: Deep learning target detection and data analysis; using the feature recognition capability of deep learning, the key features in the oil level gauge and pressure gauge images are accurately located and analyzed. The oil level monitoring specifically includes the following steps:

[0159] S111: Input oil level gauge picture;

[0160] S112: Labeling rules: labeling the oil level value, scale 0, scale 20, scale 40, scale 60, and scale 80 in the picture according to the rules;

[0161] S113: Identify the positions of the oil level frame and the scale frame through a deep learning model, and determine the coordinates of the center point of the oil level frame and the center point of the scale frame; specifically, define the coordinates of the center point of the oil level frame as (x oil ,y oil ), the coordinates of the center point of the scale frame are where i∈(0,20,40,60,80);

[0162] S114: sorting all the center points of the scale frames vertically to obtain a scale order, and generating a scale baseline function f(x) using a polynomial fitting method;

[0163] S115: Get the center point of the oil level frame (x oil ,y oil ), calculate the intersection point of its vertical projection on the scale reference line (x proj ,y proj ); the projection intersection is calculated by the scale reference line function f(x) (x proj ,y proj )=(x oil ,f(x oil ));

[0164] S116: Determine the scale range where the intersection point is located, and calculate the oil level value V using an interpolation formula. The interpolation formula is:

[0165]

[0166] Among them, lower_scale represents the oil level value of the lower scale frame of the scale range where the intersection is located, and upper_scale represents the oil level value of the upper scale frame of the scale range where the intersection is located. lower_scale Represents the y coordinate of the center point of the scale frame below the scale range where the intersection point is located, y upper_scale Represents the y-coordinate of the center point of the upper scale box of the scale range where the intersection point is located.

[0167] The pressure monitoring specifically includes the following steps:

[0168] S121: Input pressure gauge picture;

[0169] S122: Label the pointer and dial in the picture according to the labeling rules;

[0170] S123: Identify the dial center and pointer position of the pressure gauge through deep learning model and Hough circle transform, and set the dial center (c x ,c y ), pointer center point (p x ,p y );

[0171] S124: Calculate the angle θ between the pointer and the horizontal right direction. The formula is:

[0172]

[0173] Where θ is the calculated angle in degrees, (c x ,c y ) is the coordinate of the dial center, (p x ,py ) is the coordinate of the center point of the pointer;

[0174] S125: Correct the angle θ to the range [0, 360]. If θ<0, correct it to θ=θ+360.

[0175] S126: Set the effective angle range of the pressure gauge, map the angle θ to the pressure value, and use the normalization formula to convert the angle θ into the actual value P of the pressure gauge. The specific formula is:

[0176]

[0177] Where P is the pressure value calculated based on the pointer position, min_value is the minimum value of the dial, and max_value is the maximum value of the dial; θ min is the minimum value of the effective angle range of the pressure gauge, θ max It is the maximum value of the effective angle range of the pressure gauge.

[0178] Both oil level monitoring and pressure monitoring import RGB format images.

[0179] S2: Dynamic calibration: Considering the complexity of the actual application environment, by introducing an environmental perception mechanism, the image recognition threshold is dynamically adjusted and the visual detection model is optimized according to changes in environmental factors such as temperature, humidity, and light, so that the system can work stably and accurately in different environments. The specific steps include:

[0180] S21: Input data, including temperature, humidity, and light intensity values collected through environmental sensor data;

[0181] S22: Parameter definition, define image preprocessing parameters, including contrast adjustment coefficient C, which ranges from 0.5 to 2.0, brightness adjustment coefficient B, which ranges from -50 to 50; define deep learning model confidence threshold, including initial confidence threshold T0 = 0.5, dynamically adjusted confidence threshold T dynamic ;

[0182] S23: Data collection and normalization: Real-time environmental data is acquired through temperature and humidity sensors and light sensors. Temperature, humidity, and light intensity are normalized to the range [0, 1] to facilitate subsequent calculations. The normalization formula is as follows:

[0183]

[0184] Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters;

[0185] S24: Image preprocessing parameter adjustment: Dynamically adjust image preprocessing parameters according to the normalized environmental data, including the contrast adjustment coefficient C, whose formula is:

[0186] C = 1.0 + (0.5 × normalized light intensity) - (0.2 × normalized humidity)

[0187] Brightness adjustment coefficient B, its formula is

[0188] B = -50 + (100 × normalized light intensity)

[0189] S25: Dynamically update the confidence threshold of the deep learning model. The confidence threshold adjustment rule includes the following steps:

[0190] Set the weight W of environmental factors on image quality enν :

[0191] W env =0.4×normalized light intensity + 0.3×normalized humidity + 0.3×normalized temperature

[0192] Then calculate the dynamic confidence threshold T dynamic , the calculation formula is:

[0193] T dynamic =T0+(W env ×0.3)

[0194] Among them, W enν is the weight of the impact of environmental factors on image quality, T0 is the initial confidence threshold;

[0195] When the environmental conditions are poor, lower the confidence threshold to reduce the missed detection rate; when the environmental conditions are good, increase the confidence threshold T dynamic To reduce false alarm rates;

[0196] S26: Image preprocessing and model inference, image preprocessing is performed, and the original image is preprocessed according to the adjusted contrast coefficient C and brightness coefficient B, including the following steps:

[0197] To make contrast adjustments:

[0198] I contrast =C×(I original -128)+128

[0199] Among them, I original Represents the pixel value of the original image, C is the contrast adjustment coefficient, I contrast is the pixel value after contrast adjustment;

[0200] To make brightness adjustments:

[0201] I brightness =I contrast +B

[0202] Among them, I contrast The pixel value after contrast adjustment, B is the brightness adjustment coefficient, I brightness is the pixel value after adjusting the brightness;

[0203] The preprocessed images are used as input to the deep learning model;

[0204] Perform model inference:

[0205] Use the adjusted confidence threshold T dynamic Perform target detection and recognition;

[0206] If the confidence level of the test result is lower than T dynamic , then ignore the test result; otherwise, retain it and process it further.

[0207] S27: Establish a feedback mechanism, including false alarm rate monitoring and adaptive learning. The false alarm rate monitoring counts the false alarm rate in real time and analyzes it in combination with environmental data. If the false alarm rate is high, the preprocessing parameters and confidence threshold adjustment rules are further optimized; the adaptive learning associates historical environmental data with the detection results and trains a lightweight regression model to predict the optimal preprocessing parameters and confidence threshold.

[0208] S3: Intelligent Alarm and Predictive Warning: By embedding a trend prediction model and analyzing the changing trends of oil level and pressure, the system can upgrade from passive alarm to active warning. The system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses. The specific steps include:

[0209] S31: Input data, including real-time monitoring data of oil level and pressure values, historical change records of oil level and pressure, lower limits of oil level and pressure, warning time window, set time point sequence T = {t1, t2, ..., tn}, and collect corresponding oil level or pressure value sequence X = {x1, x2, ..., xn};

[0210] S32: Determine the trend prediction model and use the LSTM long short-term memory network prediction model to determine the user-set oil level safety lower limit Lmin and pressure safety lower limit Pmin; dynamically adjust the dynamic warning threshold based on the volatility of historical data;

[0211] S33: Data preprocessing, removing outliers, using interpolation or mean to fill missing values, and normalization to normalize the oil level and pressure data to the [0, 1] interval to facilitate model training. The formula is:

[0212]

[0213] Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters;

[0214] S34: Trend prediction model training, using the LSTM model for time series prediction. The input features are the normalized oil level and pressure data series; the output targets are the predicted oil level and pressure values for a period of time in the future. During the training process, the LSTM model is trained by dividing the historical data into training and test sets.

[0215] S35: Trend prediction: input current data and the most recent time series data into the LSTM model for prediction, predicting the oil level and pressure values in the future.

[0216] S36: Safety threshold judgment: if the predicted value is lower than the safety lower limit set by the user, an early warning is triggered;

[0217] The alarm type is set as level one warning or level two warning according to the distance between the predicted value and the threshold. A level one warning is triggered when the predicted value is close to the safety threshold, and a level two warning is triggered when the predicted value exceeds the safety threshold.

[0218] S4: Data storage and visual analysis, stores historical data, and performs trend analysis and visual display. By analyzing historical data, the patterns and potential problems of equipment operation can be discovered, providing a decision-making basis for equipment maintenance and management. Visual display can present monitoring data and analysis results in an intuitive way, making it easier for users to view and understand.

Claims

1. A state monitoring method integrating oil level and pressure monitoring alarm, characterized in that: The following steps are involved: S1: Deep Learning Target Detection and Data Parsing: Using deep learning’s feature recognition capabilities, we accurately locate and analyze key features in oil level gauge and pressure gauge images, providing data support for subsequent measurement and analysis. S2: Dynamic Calibration: Taking into account the complexity of actual application environments, by introducing an environmental perception mechanism, the image recognition threshold is dynamically adjusted and the visual detection model is optimized according to changes in environmental factors such as temperature, humidity, and lighting, so that the system can work stably and accurately in different environments. S3: Intelligent Alarm and Predictive Warning: By embedding a trend prediction model to analyze oil level and pressure trends, the system upgrades from passive alarms to active warnings. This system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses. S4: Data storage and visual analysis, stores historical data, and performs trend analysis and visual display. By analyzing historical data, the patterns and potential problems of equipment operation can be discovered, providing a decision-making basis for equipment maintenance and management. Visual display can present monitoring data and analysis results in an intuitive way, making it easier for users to view and understand.

2. A state monitoring method integrating oil level and pressure monitoring and alarm according to claim 1, characterized in that: Step S1 deep learning target detection and data analysis specifically includes oil level monitoring and pressure monitoring.

3. The state monitoring method integrating oil level and pressure monitoring and alarm according to claim 2 is characterized in that: The oil level monitoring specifically includes the following steps: S111: Input oil level gauge picture; S112: Labeling rules: labeling the oil level value, scale 0, scale 20, scale 40, scale 60, and scale 80 in the picture according to the rules; S113: Identify the positions of the oil level frame and the scale frame through a deep learning model, and determine the coordinates of the center point of the oil level frame and the center point of the scale frame; specifically, define the coordinates of the center point of the oil level frame as (x oil ,y oil ), the coordinates of the center point of the scale frame are where i∈(0,20,40,60,80); S114: sorting all the center points of the scale frames vertically to obtain a scale order, and generating a scale baseline function f(x) using a polynomial fitting method; S115: Get the center point of the oil level frame (x oil ,y oil ), calculate the intersection point of its vertical projection on the scale reference line (x proj ,y proj ); the projection intersection is calculated by the scale reference line function f(x) (x proj ,y proj )=(x oil ,f(x oil )); S116: Determine the scale range where the intersection point is located, and calculate the oil level value V using an interpolation formula. The interpolation formula is: Among them, lower_scale represents the oil level value of the lower scale frame of the scale range where the intersection is located, and upper_scale represents the oil level value of the upper scale frame of the scale range where the intersection is located. lower_scale Represents the y coordinate of the center point of the scale frame below the scale range where the intersection point is located, y upper_scale Represents the y-coordinate of the center point of the upper scale box of the scale range where the intersection point is located.

4. The state monitoring method integrating oil level and pressure monitoring and alarming according to claim 2 is characterized in that: The pressure monitoring specifically includes the following steps: S121: Input pressure gauge picture; S122: Label the pointer and dial in the picture according to the labeling rules; S123: Use deep learning model and Hough circle transform to identify the dial center and pointer position of the pressure gauge, and set the dial center (c x ,c y ), pointer center point (p x ,p y ); S124: Calculate the angle θ between the pointer and the horizontal right direction. The formula is: Where θ is the calculated angle in degrees, (c x ,c y ) is the coordinate of the dial center, (p x ,p y ) is the coordinate of the center point of the pointer; S125: Correct the angle θ to the range [0, 360]. If θ<0, correct it to θ=θ+360. S126: Set the effective angle range of the pressure gauge, map the angle θ to the pressure value, and use the normalization formula to convert the angle θ into the actual value P of the pressure gauge. The specific formula is: Where P is the pressure value calculated based on the pointer position, min_value is the minimum value of the dial, and max_value is the maximum value of the dial; θ min is the minimum value of the effective angle range of the pressure gauge, θ max It is the maximum value of the effective angle range of the pressure gauge.

5. The state monitoring method integrating oil level and pressure monitoring and alarming according to claim 2 is characterized in that: The oil level monitoring and pressure monitoring are both imported into RGB format images.

6. The state monitoring method integrating oil level and pressure monitoring and alarm according to claim 1 is characterized in that: Step S2 dynamic calibration specifically involves introducing an environmental perception mechanism, combining temperature, humidity, light and other sensor data, dynamically adjusting the image recognition threshold, and optimizing the visual detection model. Specifically, it includes the following steps: S21: Input data, including temperature, humidity, and light intensity values collected through environmental sensor data; S22: Parameter definition, define image preprocessing parameters, including contrast adjustment coefficient C, which ranges from 0.5 to 2.0, brightness adjustment coefficient B, which ranges from -50 to 50; define deep learning model confidence threshold, including initial confidence threshold T0 = 0.5, dynamically adjusted confidence threshold T dynamic ; S23: Data collection and normalization: Real-time environmental data is acquired through temperature and humidity sensors and light sensors. Temperature, humidity, and light intensity are normalized to the range [0, 1] to facilitate subsequent calculations. The normalization formula is as follows: Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters; S24: Image preprocessing parameter adjustment: Dynamically adjust image preprocessing parameters according to the normalized environmental data, including the contrast adjustment coefficient C, whose formula is: C = 1.0 + (0.5 × normalized light intensity) - (0.2 × normalized humidity) Brightness adjustment coefficient B, its formula is: B = -50 + (100 × normalized light intensity) S25: Dynamically update the confidence threshold of the deep learning model. The confidence threshold adjustment rule includes the following steps: Set the weight W of environmental factors on image quality enν : W env =0.4×normalized light intensity + 0.3×normalized humidity + 0.3×normalized temperature Then calculate the dynamic confidence threshold T dynamic , the calculation formula is: T dynamic =T0+(W env ×0.3) Among them, W enν is the weight of the impact of environmental factors on image quality, T0 is the initial confidence threshold; When the environmental conditions are poor, lower the confidence threshold to reduce the missed detection rate; when the environmental conditions are good, increase the confidence threshold T dynamic To reduce false alarm rates; S26: Image preprocessing and model inference, image preprocessing is performed, and the original image is preprocessed according to the adjusted contrast coefficient C and brightness coefficient B, including the following steps: To make contrast adjustments: I contrast =C×(I original -128)+128 Among them, I original Represents the pixel value of the original image, C is the contrast adjustment coefficient, I contrast is the pixel value after contrast adjustment; To make brightness adjustments: I brightness =I contrast +B Among them, I contrast The pixel value after contrast adjustment, B is the brightness adjustment coefficient, I brightness is the pixel value after adjusting the brightness; The preprocessed images are used as input to the deep learning model; Perform model inference: Use the adjusted confidence threshold T dynamic Perform target detection and recognition; If the confidence level of the test result is lower than T dynamic , then ignore the test result; otherwise, retain it and process it further. S27: Establish a feedback mechanism, including false alarm rate monitoring and adaptive learning. The false alarm rate monitoring counts the false alarm rate in real time and analyzes it in combination with environmental data. If the false alarm rate is high, the preprocessing parameters and confidence threshold adjustment rules are further optimized; the adaptive learning associates historical environmental data with the detection results and trains a lightweight regression model to predict the optimal preprocessing parameters and confidence threshold.

7. The state monitoring method integrating oil level and pressure monitoring and alarming according to claim 1 is characterized in that: Step S3: Intelligent alarm and prediction warning. By embedding a trend prediction model, it analyzes the changing trends of oil level and pressure, realizing the upgrade from passive alarm to active warning. The system can identify equipment failure risks in advance, improve operation and maintenance efficiency, and reduce potential losses. Specifically, it includes the following steps: S31: Input data, including real-time monitoring data of oil level and pressure values, historical change records of oil level and pressure, lower limits of oil level and pressure, warning time window, set time point sequence T = {t1, t2, ..., tn}, and collect corresponding oil level or pressure value sequence X = {x1, x2, ..., xn}; S32: Determine the trend prediction model and use the LSTM long short-term memory network prediction model to determine the user-set oil level safety lower limit Lmin and pressure safety lower limit Pmin; dynamically adjust the dynamic warning threshold based on the volatility of historical data; S33: Data preprocessing, removing outliers, using interpolation or mean to fill missing values, and normalization to normalize the oil level and pressure data to the [0, 1] interval to facilitate model training. The formula is: Among them, X is the original data, X min and X max The minimum and maximum values of the corresponding parameters; S34: Trend prediction model training, using the LSTM model for time series prediction. The input features are the normalized oil level and pressure data series; the output targets are the predicted oil level and pressure values for a period of time in the future. During the training process, the LSTM model is trained by dividing the historical data into training and test sets. S35: Trend prediction: input current data and the most recent time series data into the LSTM model for prediction, predicting the oil level and pressure values in the future. S36: Safety threshold judgment: if the predicted value is lower than the safety lower limit set by the user, an early warning is triggered; The alarm type is set as level one warning or level two warning according to the distance between the predicted value and the threshold. Level one warning is triggered when the predicted value is close to the safety threshold, and level two warning is triggered when the predicted value exceeds the safety threshold.

8. A state monitoring system integrating oil level and pressure monitoring and alarm, characterized in that: It includes a perception layer, a data processing layer, an analysis and decision-making layer, and an application layer. The perception layer collects images of oil level gauges and pressure gauges, uses sensors to collect temperature, humidity, and light data, and transmits data with the help of wireless communication modules. The data processing layer uses a deep learning target detection module to identify key features, an intelligent computing module to perform high-precision measurements, and a dynamic calibration module to optimize detection based on environmental data. The analysis and decision-making layer uses a trend prediction model to analyze change trends, and a threshold judgment module compares predicted values with thresholds to trigger corresponding actions. When data abnormalities occur at the application layer, the remote alarm module notifies the operation and maintenance personnel through SMS and WeChat applet push functions of abnormal information. The data storage module saves historical data, and the visual analysis module displays the monitoring results.

9. A terminal device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images