Gasoline saw overheating grading protection method and system based on exhaust temperature feedback

Abstract

The application discloses a gasoline saw superheat grading protection method and system based on exhaust temperature feedback, comprising: collecting exhaust port thermocouple signals and crankshaft phase pulse, dividing temperature waveform according to working cycle and extracting multi-dimensional feature spectrum vector; constructing a double-node thermal state space model, inputting the feature vector sequentially for time-varying parameter estimation and innovation correction, realizing internal temperature reconstruction; fusing the crankshaft speed and the feature vector for load mode identification, constructing a thermal damage risk factor and generating a continuous risk index sequence; dividing the risk interval according to a preset rule, executing multi-mode cooling intervention, and dynamically adjusting the protection strategy in combination with cycle judgment; non-invasively monitoring the health state of the sensor during the protection process, determining the degradation degree based on mean value drift, variance anomaly and autocorrelation decay constant, implementing signal correction and pushing predictive maintenance information, thereby ensuring operation safety and prolonging the service life of the equipment.

Description

Technical Field

[0001] This invention relates to the field of gasoline saw control technology, and in particular to a gasoline saw overheat protection method and system based on exhaust temperature feedback. Background Technology

[0002] Chainsaws, widely used handheld power tools in forestry logging, landscaping, and emergency rescue, are powered by small two-stroke gasoline engines. Two-stroke engines are commonly used due to their compact structure and high power-to-weight ratio. However, during operation, they perform combustion work with each revolution, resulting in concentrated heat load. Furthermore, during scavenging, some of the fresh air-fuel mixture is directly short-circuited to the exhaust manifold, causing the combustion chamber and exhaust port area to be under prolonged high-temperature alternating stress conditions. In actual operation, operators often experience a continuous deterioration of the engine's thermal condition due to sawing hardwood, chain dulling, or prolonged continuous full-load cutting. When the cylinder wall temperature exceeds the material's allowable limit, irreversible thermal damage such as piston scoring, piston ring sticking, cylinder wall scratches, and even complete engine seizure can easily occur, severely impacting the machine's lifespan and operational safety.

[0003] To address the aforementioned overheating issues, existing protection solutions typically install temperature sensors in locations such as the cylinder block or spark plug seat. When the detected temperature exceeds a single fixed threshold, forced ignition cutoff or engine shutdown is executed. This approach suffers from the following technical drawbacks: The thermal response at the cylinder block temperature sensor exhibits significant lag; by the time the cylinder temperature sensor detects a dangerous temperature, the piston ring area near the exhaust port has already far exceeded safety limits, resulting in delayed protection intervention. A single fixed threshold cannot accommodate the permissible temperature differences under various operating conditions. The safe temperature boundaries for high-speed, heavy-load conditions differ by tens of degrees Celsius from those for low-speed, light-load conditions. Using the same threshold can easily lead to protection lag under high loads and false triggering of power limiting under low loads. Protection actions primarily rely on sudden engine shutdown or forced shutdown. Sudden stops during sawing operations can easily cause the saw chain to jam in the saw kerf, leading to operational safety hazards such as saw rebound and affecting operational continuity. Furthermore, the exhaust temperature of a two-stroke engine fluctuates dramatically between cycles due to the scavenging short circuit. Simple threshold comparisons are prone to frequent false triggering due to single-cycle temperature spikes, lacking effective suppression of sensor signal noise and transient interference. Therefore, there is an urgent need for an overheat protection method that is fast-responding, progressively graded, adaptable to operating conditions, and highly resistant to interference, so as to balance engine safety and operational continuity. Summary of the Invention

[0004] This invention overcomes the shortcomings of the prior art and provides a method and system for graded overheat protection of gasoline saws based on exhaust temperature feedback. Its important purpose is to improve the overheat response speed and enhance the adaptability of operating conditions, thereby reducing the risk of false triggering of protection.

[0005] To achieve the above objectives, the first aspect of the present invention provides a method for graded overheat protection of a gasoline saw based on exhaust temperature feedback, comprising: Collect fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals. Divide the window according to the working cycle and perform dynamic segmentation of the exhaust temperature waveform within the cycle. Simultaneously extract the working cycle feature parameters to construct a multi-dimensional feature spectrum vector. A two-node thermal state space model is constructed. The multidimensional feature spectrum vector of each working cycle is used as the sequential observation input to estimate the time-varying parameters of thermal state transfer between cycles. The prior state is corrected by calculating the observation information and the internal temperature is reconstructed to obtain the internal predicted temperature. The instantaneous crankshaft speed sequence is obtained, and multi-parameter phase space mapping of load modes is performed by combining multi-dimensional feature spectrum vectors. Load mode recognition is performed in the mapping space using clustering algorithms. Based on real-time clustering results and internal predicted temperature, thermal damage risk factors are constructed to classify and quantify thermal damage risk, and a continuous thermal damage risk index sequence is generated. Based on the preset overheating risk judgment rules, the continuous thermal damage risk index sequence is divided into different risk intervals to obtain the overheating risk sequence that evolves over time. Based on the overheating risk sequence, multi-mode cooling intervention is performed to generate a continuous protection action plan for execution. During the execution, cyclic risk judgment is performed to adjust the protection action plan. During the continuous operation of the overheat protection, the long-term health status of the exhaust temperature sensor is monitored non-invasively. The degree of degradation is determined by the mean drift, the percentage increase in variance, and the autocorrelation decay time constant. Degradation correction is performed based on the degree of degradation, and predictive maintenance information is generated and pushed out.

[0006] In this scheme, the fast-response thermocouple signal from the exhaust port of the target gasoline saw and the crankshaft phase pulse signal are collected. Windows are divided according to the working cycle, and dynamic segmentation of the exhaust temperature waveform within the cycle is performed. Simultaneously, working cycle feature parameters are extracted to construct a multi-dimensional feature spectrum vector. Specifically, this includes: Collect fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals. Input the fast-response thermocouple signals to an analog-to-digital converter to convert them into digital voltage sequences at a preset sampling rate. Based on the thermocouple indexing characteristics, map the digital voltage sequences into exhaust temperature sampling sequences. Synchronously convert the crankshaft phase pulse signal into an edge-triggered sequence and record the pulse interval between adjacent top dead center points. Based on the pulse count, divide the exhaust temperature sampling sequence into several independent working cycle windows according to the preset crankshaft rotation angle. For the exhaust temperature sampling sequence within each working cycle window, the local standard deviation of each sampling point is calculated by sliding the fixed narrow window width point by point as the instantaneous noise base estimate. The noise base estimate is then arranged along the time axis to form the noise base curve within the current cycle. Based on the noise floor curve, an exhaust pulse determination rule is established, the first-order forward difference of the exhaust temperature sampling sequence is calculated, and the exhaust pulse rise start point, exhaust pulse peak point and exhaust pulse end point are marked using the exhaust pulse determination rule. Starting from the end point of the exhaust pulse, continue tracking to the lowest temperature region where the exhaust temperature fluctuation is less than the first threshold and the first-order difference fluctuates narrowly near zero. Search for the minimum sample value in the lowest temperature region as the valley temperature point after scavenging, and complete the calibration of the exhaust pulse peak point, valley temperature point, rising start point and end point within the current working cycle window. Based on the calibration results, the peak temperature, valley temperature, maximum gradient of rising and falling edges, peak half width, pulse area integral, and asymmetry ratio are extracted and combined in sequence to form the feature vector sequence of the current working cycle. After performing directional adaptive filtering on the feature vector sequences of several consecutive working cycles, a multidimensional feature spectrum vector is obtained.

[0007] In this scheme, the estimation of time-varying parameters for inter-cycle thermal state transfer involves calculating observational information to correct prior states and reconstructing the internal temperature to obtain the predicted internal temperature. Specifically, this includes: A two-node thermal state space model is constructed with cylinder wall high-temperature nodes and exhaust port wall nodes as state variables. The cylinder wall high-temperature nodes correspond to the cylinder wall region near the exhaust port, and the exhaust port wall nodes correspond to the inner wall region of the exhaust port. Each node has independent heat capacity and temperature state. The nodes are coupled through heat conduction terms and have convective heat transfer paths with the external environment. Taking the first working cycle after the cold start of the gasoline saw as the starting point, the initial temperatures of the high-temperature nodes of the cylinder wall and the exhaust port wall nodes are set to the measured values ​​of the ambient temperature sensor. The initial state covariance matrix is ​​set as a diagonal matrix, and the values ​​of the diagonal elements represent the degree of uncertainty of the initial temperature estimation. Before each state prediction, the convective heat transfer coefficient in the state transition matrix is ​​updated in real time based on the engine speed and throttle opening of the current working cycle. The multidimensional feature spectrum vector of the current working cycle is used as the sequential observation input, and the peak temperature and valley temperature are used as the dual measurement outputs to establish the peak temperature measurement equation and the valley temperature measurement equation. After completing the state transition matrix update and measurement equation establishment, the prior state prediction is performed. The posterior temperature estimates of the cylinder wall high temperature node and exhaust port wall node output from the previous working cycle are used as the initial state of this cycle. They are substituted into the updated state transition matrix to obtain the prior temperature estimate and the prior state covariance matrix. Substitute the prior temperature estimate into the peak temperature measurement equation and the valley temperature measurement equation to calculate the predicted peak temperature and the predicted valley temperature. Calculate the difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, respectively, to form the observation information vector. The deviation amplitude between the peak temperature and the valley temperature in the multidimensional feature spectrum vector before and after directional adaptive filtering is calculated. The measurement noise covariance matrix is ​​adjusted according to the deviation amplitude. The Kalman gain is calculated using the adjusted measurement noise covariance. The prior temperature estimate is corrected by weighting the observation information vector with the Kalman gain to obtain the posterior temperature estimate. The posterior temperature estimate of the high temperature node of the cylinder wall is taken as the internal predicted temperature.

[0008] In this scheme, the multi-parameter phase space mapping of load modes, the load mode identification in the mapping space using a clustering algorithm, and the construction of thermal damage risk factors based on real-time clustering results and internal predicted temperature for thermal damage risk classification and quantification specifically include: Extract the instantaneous crankshaft speed sequence of each working cycle from the pulse interval sequence of the crankshaft position sensor, which shows the instantaneous speed change with the crankshaft angle. Calculate the difference between the highest and lowest instantaneous speeds within the cycle to obtain the peak-to-peak speed. Perform a Fourier transform on the instantaneous crankshaft speed sequence and calculate the proportion of half-order harmonic energy. Extract the crankshaft angle phase at the moment when the exhaust peak temperature occurs and the crankshaft angle phase at the moment when the speed peak occurs, and calculate the difference between the two as the phase offset. The moment when the exhaust peak temperature occurs is determined by the calibration result of the exhaust pulse peak point corresponding to the multidimensional feature spectrum vector. The peak half-width and the maximum gradient at the falling edge of the current working cycle are extracted from the multidimensional feature spectrum vector. The peak half-width and the maximum gradient at the falling edge are multiplied to obtain the single-cycle exhaust thermal impulse characterization value. This value, together with the peak-to-peak speed, the proportion of half-order harmonic energy, and the phase offset, constitutes a four-dimensional load feature vector, which serves as the real-time mapping point of the current working cycle in the multi-parameter phase space. Four cluster centers—preset empty mode, light load cutting mode, heavy load cutting mode, and cam saw overload mode—are used as initial benchmarks. The Euclidean distance between the current four-dimensional load feature vector and the four preset cluster centers is calculated using the nearest neighbor classification algorithm. The load mode corresponding to the preset cluster center with the smallest Euclidean distance is selected as the classification result of the current working cycle, thus obtaining the load mode identification information.

[0009] Based on the load mode recognition information, the upper limit value of cylinder wall temperature corresponding to the current load mode is retrieved from the preset safety limit parameter group, and the internal estimated temperature is divided by the upper limit value of cylinder wall temperature to obtain the temperature margin ratio. Meanwhile, the diagonal elements of the posterior variance corresponding to the high-temperature nodes of the cylinder wall are extracted from the posterior state covariance matrix of the two-node thermal state space model. The estimated normalized variance value is obtained by normalizing the diagonal elements corresponding to the initial covariance matrix of the cold start as a benchmark. The thermal damage risk factor for the current work cycle is obtained by multiplying the temperature margin ratio by the estimated variance normalized value. The thermal damage risk factors for multiple consecutive work cycles are then subjected to sliding window mean filtering to generate a continuous thermal damage risk index sequence.

[0010] In this scheme, the multi-mode cooling intervention based on the overheating risk sequence generates a continuous protection action plan for execution, and during execution, a cyclical risk assessment is performed to adjust the protection action plan. Specifically, this includes: The continuous thermal damage risk index sequence is divided into four risk intervals in ascending order: normal interval, warning interval, restricted interval, and emergency interval. A transition overlap zone is set between adjacent intervals, and the boundary threshold of the interval division is dynamically shifted based on the current load mode identification result. Each index value in the continuous thermal damage risk index sequence is mapped to the corresponding risk interval to form an overheating risk sequence that evolves over time. Each element in the overheating risk sequence contains the risk interval to which the current work cycle belongs and the relative position ratio of the risk index within its respective interval. Based on the risk zone attribution and relative position ratio of each working cycle in the overheat risk sequence, multi-mode cooling intervention is executed to generate a continuous protection action plan; if the current element belongs to the normal range, the original engine control parameters are maintained, the injection pulse width is calculated according to the normal power air-fuel ratio, and the ignition advance angle is kept at the original engine calibration value. If the current element belongs to the warning range, an additional injection pulse width is superimposed on the basic injection pulse width. The amplitude of the additional injection pulse width is linearly positively correlated with the relative position ratio of the risk index within the warning range. The ignition advance angle is delayed and the delay amplitude is linearly correlated with the relative position ratio. The maximum allowable speed of the engine is limited to the preset intermediate speed level. If the current element belongs to the restricted range, the throttle opening will be continuously reduced along the preset decay curve, the fuel injection mode will be changed to periodic intermittent injection, and the non-fuel injection working cycle will retain ignition to maintain the crankshaft rotational inertia, and the engine speed will be maintained at the preset idle speed level to drive the cooling fan to operate. If the current element belongs to the emergency zone, the throttle will be completely closed, cutting off the fuel injection signal and ignition signal. After the engine stops, the operator must manually reset it before it can be restarted. During the execution of the protection action plan, each work cycle will re-determine the cycle risk based on the newly generated thermal damage risk index and dynamically adjust the protection action plan parameters.

[0011] In this solution, the long-term health status of the exhaust temperature sensor is monitored non-invasively. The degree of degradation is determined by the mean drift, the percentage increase in variance abnormality, and the autocorrelation decay time constant. Degradation correction is performed based on the degree of degradation, and predictive maintenance information is generated and pushed to relevant departments. Specifically, this includes: The difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, output by the two-node thermal state space model in each working cycle, constitute an information sequence, which serves as the data source for sensor health status monitoring. Sliding window recursive statistics are performed on the information sequence of a continuous preset number of working cycles. The smoothed values ​​of peak temperature information and valley temperature information are calculated using the exponentially weighted moving average algorithm, and the larger of the two absolute values ​​is taken as the mean drift. The larger of the variance of peak temperature information and the variance of valley temperature information within the window is divided by the factory calibration reference variance to obtain the variance abnormality increase ratio. The first-order autoregressive model coefficients of the peak temperature information sequence within the sliding window were fitted using the Yule-Walker equation. The autocorrelation coefficient of the lag step was extracted and substituted into the exponential decay model. The number of lag steps corresponding to the decay of the normalized autocorrelation function to the preset threshold was calculated and converted into crankshaft rotation angle as the autocorrelation decay time constant. The mean drift, variance aberration increase rate, and autocorrelation decay time constant are compared with preset slight degradation thresholds and moderate degradation thresholds, respectively. If at least one indicator exceeds the slight degradation threshold and no indicator reaches the moderate degradation threshold, it is rated as slight degradation. If at least one indicator exceeds the moderate degradation threshold, it is rated as moderate degradation. When the degradation is assessed as slight, the measured signal is maintained to participate in the closed-loop correction. The multiplication coefficient of the estimated variance normalization value used in calculating the thermal damage risk factor is increased by a preset ratio to make the overheat protection decision more conservative. Warning information containing degradation index values ​​and mild maintenance recommendations is generated and pushed out. When the degradation is rated as moderate, a non-disruptive switch from closed-loop calibration to open-loop prediction is initiated. This causes the inverse matrix norm of the measurement noise covariance matrix to decrease smoothly over multiple consecutive working cycles according to an exponential decay function, so that the Kalman gain decays synchronously to zero. After the switch, a pure open-loop prediction is performed by a two-node thermal state-space model. The safe upper limit of cylinder wall temperature is multiplied by a margin factor, and maintenance information including sensor replacement recommendations is generated and pushed out.

[0012] A second aspect of the present invention provides a gasoline saw overheating graded protection system based on exhaust temperature feedback. The system includes: a memory, a processor, and a communication interface. The memory contains a gasoline saw overheating graded protection method program based on exhaust temperature feedback. When the processor executes the gasoline saw overheating graded protection method program based on exhaust temperature feedback, it implements the steps of the gasoline saw overheating graded protection method based on exhaust temperature feedback as described in any of the preceding claims. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in the embodiments or examples of the present invention, the drawings used in the embodiments or examples will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained according to these drawings without creative effort.

[0014] Figure 1 A flowchart of a gasoline saw overheating graded protection method based on exhaust temperature feedback provided in an embodiment of the present invention; Figure 2 A flowchart of a method for diagnosing the degradation of an exhaust temperature sensor based on statistical feature analysis of a novel information sequence, provided in an embodiment of the present invention; Figure 3 A block diagram of a gasoline saw overheat protection system based on exhaust temperature feedback provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0016] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0017] Figure 1 A flowchart of a gasoline saw overheating graded protection method based on exhaust temperature feedback provided in an embodiment of the present invention; like Figure 1 As shown, the present invention provides a flowchart of a graded overheat protection method for gasoline saws based on exhaust temperature feedback, comprising: Collect fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals. Divide the window according to the working cycle and perform dynamic segmentation of the exhaust temperature waveform within the cycle. Simultaneously extract the working cycle feature parameters to construct a multi-dimensional feature spectrum vector. A two-node thermal state space model is constructed. The multidimensional feature spectrum vector of each working cycle is used as the sequential observation input to estimate the time-varying parameters of thermal state transfer between cycles. The prior state is corrected by calculating the observation information and the internal temperature is reconstructed to obtain the internal predicted temperature. The instantaneous crankshaft speed sequence is obtained, and multi-parameter phase space mapping of load modes is performed by combining multi-dimensional feature spectrum vectors. Load mode recognition is performed in the mapping space using clustering algorithms. Based on real-time clustering results and internal predicted temperature, thermal damage risk factors are constructed to classify and quantify thermal damage risk, and a continuous thermal damage risk index sequence is generated. Based on the preset overheating risk judgment rules, the continuous thermal damage risk index sequence is divided into different risk intervals to obtain the overheating risk sequence that evolves over time. Based on the overheating risk sequence, multi-mode cooling intervention is performed to generate a continuous protection action plan for execution. During the execution, cyclic risk judgment is performed to adjust the protection action plan. During the continuous operation of the overheat protection, the long-term health status of the exhaust temperature sensor is monitored non-invasively. The degree of degradation is determined by the mean drift, the percentage increase in variance, and the autocorrelation decay time constant. Degradation correction is performed based on the degree of degradation, and predictive maintenance information is generated and pushed out.

[0018] Furthermore, in a preferred embodiment of the present invention, the step of acquiring the fast-response thermocouple signal from the exhaust port of the target gasoline saw and the crankshaft phase pulse signal, dividing the window according to the working cycle and performing dynamic segmentation of the exhaust temperature waveform within the cycle, and simultaneously extracting the working cycle feature parameters to construct a multi-dimensional feature spectrum vector, specifically includes: Collect fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals. Input the fast-response thermocouple signals to an analog-to-digital converter to convert them into digital voltage sequences at a preset sampling rate. Based on the thermocouple indexing characteristics, map the digital voltage sequences into exhaust temperature sampling sequences. Synchronously convert the crankshaft phase pulse signal into an edge-triggered sequence and record the pulse interval between adjacent top dead center points. Based on the pulse count, divide the exhaust temperature sampling sequence into several independent working cycle windows according to the preset crankshaft rotation angle. For the exhaust temperature sampling sequence within each working cycle window, the local standard deviation of each sampling point is calculated by sliding the fixed narrow window width point by point as the instantaneous noise base estimate. The noise base estimate is then arranged along the time axis to form the noise base curve within the current cycle. Based on the noise floor curve, an exhaust pulse determination rule is established, the first-order forward difference of the exhaust temperature sampling sequence is calculated, and the exhaust pulse rise start point, exhaust pulse peak point and exhaust pulse end point are marked using the exhaust pulse determination rule. Starting from the end point of the exhaust pulse, continue tracking to the lowest temperature region where the exhaust temperature fluctuation is less than the first threshold and the first-order difference fluctuates narrowly near zero. Search for the minimum sample value in the lowest temperature region as the valley temperature point after scavenging, and complete the calibration of the exhaust pulse peak point, valley temperature point, rising start point and end point within the current working cycle window. Based on the calibration results, the peak temperature, valley temperature, maximum gradient of rising and falling edges, peak half width, pulse area integral, and asymmetry ratio are extracted and combined in sequence to form the feature vector sequence of the current working cycle. After performing directional adaptive filtering on the feature vector sequences of several consecutive working cycles, a multidimensional feature spectrum vector is obtained.

[0019] It should be noted that the acquisition and preprocessing of the exhaust temperature signal directly determines the response speed and anti-interference capability of the subsequent overheat protection strategy. During the scavenging process of a two-stroke gasoline saw, there is a short circuit phenomenon of unburned mixture, and the thermocouple at the exhaust port will intermittently come into contact with the low-temperature airflow. This results in the original signal being superimposed with periodic spike pulses and wideband random noise synchronized with the power stroke. If conventional low-pass filtering or simple mean smoothing is used directly, it will either fail to effectively suppress spike interference or smooth out the actual temperature rise as well, thus losing the inherent advantage of exhaust temperature in rapidly responding to the in-cylinder thermal state.

[0020] To address this, this scheme first performs analog-to-digital conversion on the thermocouple signal at a sampling rate higher than the engine's highest fundamental frequency. Simultaneously, it acquires crankshaft phase pulse signals and divides the continuous exhaust temperature data stream into independent working cycle windows at 720-degree crankshaft rotation angles. This ensures that all subsequent feature extraction operations use the complete cyclic thermodynamic process as the basic analysis unit, rather than unstructured data extraction based on absolute time. Based on this, the local standard deviation is calculated point-by-point for the sampling sequence within each window using a fixed narrow window width. These local standard deviations are then arranged along the time axis to form the noise floor curve for the current cycle. This curve adaptively changes with signal fluctuations, automatically increasing in the scavenging peak region and automatically decreasing in the stable temperature region, providing a dynamic judgment benchmark that matches the signal's own noise level for subsequent pulse segmentation. Based on the noise floor curve, exhaust pulse judgment rules are established. The first-order forward difference of the sampling sequence is calculated and compared point-by-point with the noise floor curve. The starting point, peak point, and end point of the exhaust pulse are calibrated using the moment when the gradient exceeds a preset multiple of the noise floor. Unlike fixed threshold triggering methods, this adaptive gradient threshold segmentation method ensures that the pulse calibration phase does not drift with changes in speed and load, accurately capturing the complete start and end boundaries of the pulse under various operating conditions. After calibrating the pulse end point, it continues to track to the lowest temperature region where the exhaust temperature fluctuation tends to level off and searches for the valley temperature point, decoupling and extracting the residual temperature characteristics of the scavenging process from the main characteristics of the exhaust pulse. Based on the calibration results, seven feature parameters are simultaneously extracted from each cycle: peak temperature, valley temperature, maximum gradient of rising and falling edges, peak half-width, pulse area integral, and asymmetry ratio, forming a multi-dimensional feature spectrum vector. This comprehensively characterizes the transient physical features of the exhaust temperature waveform, elevating the one-dimensional temperature signal into a multi-dimensional thermodynamic observation space with high information content. Directional adaptive filtering is performed on the feature vector sequence of multiple consecutive cycles. The jump amplitude between adjacent cycles is constrained by the upper limit of the physical change rate of each feature parameter, and the moving window value is output. This retains the true thermal state step change while filtering out residual random fluctuations, providing a stable and reliable data foundation for subsequent thermal state estimation and overheating graded protection. Through the above method, this step fully analyzes and utilizes the transient temperature information of the cycle stage that is smoothed out in the traditional solution, and solves the contradiction between the violent fluctuation of the two-stroke exhaust temperature signal caused by the scavenging short circuit and the reliability of protection triggering from the source of the signal.

[0021] Furthermore, in a preferred embodiment of the present invention, the time-varying parameter estimation for inter-cycle thermal state transfer, which involves calculating observed information to correct prior states and reconstructing the internal temperature to obtain the predicted internal temperature, specifically includes: A two-node thermal state space model is constructed with cylinder wall high-temperature nodes and exhaust port wall nodes as state variables. The cylinder wall high-temperature nodes correspond to the cylinder wall region near the exhaust port, and the exhaust port wall nodes correspond to the inner wall region of the exhaust port. Each node has independent heat capacity and temperature state. The nodes are coupled through heat conduction terms and have convective heat transfer paths with the external environment. Taking the first working cycle after the cold start of the gasoline saw as the starting point, the initial temperatures of the high-temperature nodes of the cylinder wall and the exhaust port wall nodes are set to the measured values ​​of the ambient temperature sensor. The initial state covariance matrix is ​​set as a diagonal matrix, and the values ​​of the diagonal elements represent the degree of uncertainty of the initial temperature estimation. Before each state prediction, the convective heat transfer coefficient in the state transition matrix is ​​updated in real time based on the engine speed and throttle opening of the current working cycle. The multidimensional feature spectrum vector of the current working cycle is used as the sequential observation input, and the peak temperature and valley temperature are used as the dual measurement outputs to establish the peak temperature measurement equation and the valley temperature measurement equation. After completing the state transition matrix update and measurement equation establishment, the prior state prediction is performed. The posterior temperature estimates of the cylinder wall high temperature node and exhaust port wall node output from the previous working cycle are used as the initial state of this cycle. They are substituted into the updated state transition matrix to obtain the prior temperature estimate and the prior state covariance matrix. Substitute the prior temperature estimate into the peak temperature measurement equation and the valley temperature measurement equation to calculate the predicted peak temperature and the predicted valley temperature. Calculate the difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, respectively, to form the observation information vector. The deviation amplitude between the peak temperature and the valley temperature in the multidimensional feature spectrum vector before and after directional adaptive filtering is calculated. The measurement noise covariance matrix is ​​adjusted according to the deviation amplitude. The Kalman gain is calculated using the adjusted measurement noise covariance. The prior temperature estimate is corrected by weighting the observation information vector with the Kalman gain to obtain the posterior temperature estimate. The posterior temperature estimate of the high temperature node of the cylinder wall is taken as the internal predicted temperature.

[0022] It should be noted that the reason this solution constructs a two-node thermal state space model instead of directly using exhaust temperature as the basis for overheating judgment is because the exhaust port thermocouple measures the temperature of the airflow passing through the exhaust passage, not the cylinder wall temperature bearing the highest thermal load. In a two-stroke gasoline saw, the exhaust temperature fluctuates drastically with the exhaust pulses during each working cycle. A single-point temperature value cannot accurately reflect the true thermal state of critical nodes on the cylinder wall. If the peak or average exhaust temperature is used directly for protection decisions, either signal fluctuations will lead to false triggering of the protection, or the lag in heat conduction path will cause protection delays.

[0023] To address this, this scheme models the cylinder wall and exhaust manifold wall near the exhaust port as two centralized heat capacity nodes with independent heat capacity and temperature. These nodes are thermally coupled and each exchanges heat with the flowing gas and the external environment via convection. This allows for a relatively low computational complexity in approximating the complete heat transfer process from the cylinder wall to the exhaust manifold and then to the thermocouple measurement point. During the first working cycle after cold start, the initial temperatures of both nodes are set to the measured values ​​from the ambient temperature sensor. The initial state covariance matrix is ​​a diagonal matrix, and the values ​​of the diagonal elements reflect the uncertainty of the initial temperature estimate under cold start conditions, providing a reasonable prior starting point for recursive correction in subsequent cycles. Before predicting the state of each working cycle, the convective heat transfer coefficient in the state transition matrix is ​​updated in real time based on the engine speed and throttle opening of the current cycle. The engine speed determines the scavenging mass flow rate and the air velocity driven by the cooling fan, thus affecting the intensity of convective heat transfer between the node and the external environment. The throttle opening determines the in-cylinder charge and combustion heat release, which in turn affects the heat input power of the cylinder wall node. This time-varying parameter model update mechanism enables the state estimator to adjust its time constant and gain in real time according to the changes in engine operating conditions, avoiding the estimation inaccuracy problem that traditional fixed parameter models are prone to in a wide speed range.

[0024] At the observational level, the peak and valley temperatures in the multidimensional feature spectral vector are simultaneously used as dual measurement outputs. The peak temperature reflects the highest temperature of the exhaust gas after heat exchange between the high-temperature combustion gas and the wall, while the valley temperature reflects the lowest temperature of the residual gas in the exhaust manifold after scavenging. These two temperatures provide observational information about the internal node temperatures from both high-temperature and low-temperature perspectives. Using the peak and valley temperatures together to correct the prior state provides a more comprehensive constraint on the estimator's correction direction and reduces the influence of local airflow disturbances on a single measurement. After obtaining the prior temperature estimate, it is substituted into the measurement equation to calculate the predicted peak and valley temperatures. Subtracting these from the measured values ​​constitutes the observational innovation vector, which comprehensively reflects the deviation between the model prediction and the actual thermal state. Furthermore, the measurement noise covariance matrix is ​​adjusted online using the deviation amplitude of the multidimensional feature spectral vector before and after directional adaptive filtering. When the exhaust temperature signal fluctuates drastically due to scavenging short circuit, the deviation amplitude before and after filtering increases, and the measurement noise covariance increases accordingly, causing the Kalman gain to automatically decrease. The estimator assigns a lower confidence weight to the measured value and relies more on model prediction. Conversely, when the signal is stable, the deviation amplitude before and after filtering decreases, and the Kalman gain increases, causing the estimator to converge to the measured value faster. Through the above adaptive adjustment mechanism, this step infers the cylinder wall critical point temperature, which cannot be directly measured, from the instantaneous characteristics of the exhaust temperature waveform. Moreover, the estimator has a built-in one-step prediction capability, allowing overheating trends to be captured several working cycles before the cylinder block temperature measurement scheme senses them, fundamentally overcoming the protection delay problem caused by the lag in heat conduction in traditional schemes.

[0025] It is worth mentioning that the state vector of the two-node thermal state-space model is defined as: ; In the formula, For the first The state vector of each work cycle This refers to the high-temperature node temperature of the cylinder wall. This refers to the temperature at the joint of the exhaust duct wall. This is the work cycle number.

[0026] Therefore, the state transition equation can be expressed as: ; In the formula, Here is the state transition matrix. For the input matrix, To be determined by engine speed Throttle opening The constructed working condition input vector, For process noise that follows a zero-mean normal distribution, Let be the process noise covariance matrix.

[0027] The equation for measuring the peak temperature is: ; In the formula, For the first The measured peak exhaust temperature for each working cycle. This refers to the high-temperature node temperature of the cylinder wall. This refers to the temperature at the joint of the exhaust duct wall. The scavenging mass flow rate is used as the dilution weight of the exhaust gas, and its value ranges from zero to one. It is the dynamic response factor of the thermocouple and is updated as the engine speed changes. The peak temperature measurement noise follows a zero-mean normal distribution. The peak temperature measurement equation expresses the peak airflow temperature at the exhaust port as a weighted mixture of the cylinder wall high-temperature node temperature and the combustion exhaust gas temperature. The combustion exhaust gas temperature is approximately characterized by the product of the exhaust port wall node temperature and the thermocouple dynamic response factor. The mixing weight is determined by the degree of dilution of the exhaust gas by the scavenging mass flow rate.

[0028] The equation for measuring valley temperature is: ; In the formula, For the first The measured exhaust valley temperature for each working cycle. The mixing ratio of scavenging gas charge to residual exhaust gas, with a value ranging from zero to one. The valley temperature measurement noise follows a zero-mean normal distribution. The valley temperature measurement equation expresses the gas temperature in the exhaust manifold after scavenging as a weighted mixture of the high-temperature node temperature of the cylinder wall and the node temperature of the exhaust manifold wall. The low-temperature gas brought by the scavenging charge has a dilution effect on the cylinder wall node temperature, while the residual exhaust gas carries information about the exhaust manifold wall node temperature. The mixing weight is determined by the ratio of the scavenging charge to the residual exhaust gas. The thermocouple dynamic response factor... They are also included in the weighting to reflect the sensor's imperfect tracking of temperature changes.

[0029] Furthermore, in a preferred embodiment of the present invention, the step of performing multi-parameter phase space mapping of load modes, identifying load modes in the mapping space using a clustering algorithm, and constructing thermal damage risk factors based on real-time clustering results and internally estimated temperatures for thermal damage risk classification and quantification specifically includes: Extract the instantaneous crankshaft speed sequence of each working cycle from the pulse interval sequence of the crankshaft position sensor, which shows the instantaneous speed change with the crankshaft angle. Calculate the difference between the highest and lowest instantaneous speeds within the cycle to obtain the peak-to-peak speed. Perform a Fourier transform on the instantaneous crankshaft speed sequence and calculate the proportion of half-order harmonic energy. Extract the crankshaft angle phase at the moment when the exhaust peak temperature occurs and the crankshaft angle phase at the moment when the speed peak occurs, and calculate the difference between the two as the phase offset. The moment when the exhaust peak temperature occurs is determined by the calibration result of the exhaust pulse peak point corresponding to the multidimensional feature spectrum vector. The peak half-width and the maximum gradient at the falling edge of the current working cycle are extracted from the multidimensional feature spectrum vector. The peak half-width and the maximum gradient at the falling edge are multiplied to obtain the single-cycle exhaust thermal impulse characterization value. This value, together with the peak-to-peak speed, the proportion of half-order harmonic energy, and the phase offset, constitutes a four-dimensional load feature vector, which serves as the real-time mapping point of the current working cycle in the multi-parameter phase space. Four cluster centers—preset empty mode, light load cutting mode, heavy load cutting mode, and cam saw overload mode—are used as initial benchmarks. The Euclidean distance between the current four-dimensional load feature vector and the four preset cluster centers is calculated using the nearest neighbor classification algorithm. The load mode corresponding to the preset cluster center with the smallest Euclidean distance is selected as the classification result of the current working cycle, thus obtaining the load mode identification information.

[0030] Based on the load mode recognition information, the upper limit value of cylinder wall temperature corresponding to the current load mode is retrieved from the preset safety limit parameter group, and the internal estimated temperature is divided by the upper limit value of cylinder wall temperature to obtain the temperature margin ratio. Meanwhile, the diagonal elements of the posterior variance corresponding to the high-temperature nodes of the cylinder wall are extracted from the posterior state covariance matrix of the two-node thermal state space model. The estimated normalized variance value is obtained by normalizing the diagonal elements corresponding to the initial covariance matrix of the cold start as a benchmark. The thermal damage risk factor for the current work cycle is obtained by multiplying the temperature margin ratio by the estimated variance normalized value. The thermal damage risk factors for multiple consecutive work cycles are then subjected to sliding window mean filtering to generate a continuous thermal damage risk index sequence.

[0031] It should be noted that the multi-parameter phase space mapping of the load mode is introduced because the causes of overheating of chainsaws in actual operation are complex. The proportion of in-cylinder combustion heat release and sawing friction heat differs significantly under different load modes. A single temperature threshold cannot distinguish between these two heat source paths, which can easily lead to a mismatch in the sensitivity of the protection strategy. Furthermore, the instantaneous crankshaft speed sequence, which varies with the crankshaft angle within each working cycle, is extracted from the pulse interval sequence of the crankshaft position sensor. The difference between the highest and lowest instantaneous speeds within the cycle is calculated to obtain the peak-to-peak speed. This parameter directly reflects the instantaneous impact intensity of the sawing resistance on the crankshaft rotation. Under heavy load or jammed saw conditions, the saw chain is obstructed, leading to increased speed fluctuations and a corresponding increase in the peak-to-peak speed. Simultaneously, a Fourier transform is performed on the instantaneous crankshaft speed sequence, and the proportion of half-harmonic energy is calculated. The half-harmonic is a characteristic frequency component generated by a two-stroke engine once per revolution. Its proportion in the total harmonic energy is highly sensitive to load changes. Under light load and no-load conditions, the proportion of half-harmonic energy is low, while under heavy load conditions, the increased torque difference between the power stroke and the compression stroke significantly enhances the half-harmonic component.

[0032] Further, the crankshaft angle phase difference between the peak exhaust temperature and the peak engine speed is extracted as the phase shift. This phase shift reflects the temporal coupling relationship between the peak combustion heat release and the peak mechanical load. The combustion rate and torque response characteristics differ under different load modes, and the phase shift exhibits a distinguishable distribution. The peak half-width and the maximum gradient at the falling edge are extracted from the multidimensional feature spectrum vector and multiplied to obtain the single-cycle exhaust thermal impulse characterization value. This product combines the high-temperature duration of the exhaust pulse with the cooling rate into a single thermal impulse metric. Under the saw overload mode with a high proportion of frictional heat, the exhaust thermal impulse characterization value is significantly different from that of the pure combustion overload mode, thus providing an effective observation dimension for distinguishing heat source types. The above four parameters constitute a four-dimensional load feature vector, which is equivalent to mapping the load state of each working cycle to a real-time position in the multi-parameter phase space. Different load modes appear as distinguishable point clusters in this space.

[0033] Based on this, the nearest neighbor classification algorithm is used to calculate the Euclidean distance between the current four-dimensional load feature vector and four preset cluster centers, and the one with the smallest distance is selected to output the load pattern recognition information. The four preset cluster centers are obtained by collecting samples and calculating the mean values ​​under four known working conditions: no load, light load cutting, heavy load cutting, and saw overload through offline bench tests. The online recognition process has low computational load and high real-time performance. After obtaining the load pattern recognition information, the upper limit value of cylinder wall temperature corresponding to the current load pattern is retrieved from the preset safety limit parameter group. The internal estimated temperature is divided by this upper limit value to obtain the temperature margin ratio. At the same time, the diagonal elements of the posterior variance corresponding to the high temperature node of the cylinder wall are extracted from the posterior state covariance matrix of the two-node thermal state space model and normalized to obtain the estimated variance normalized value. This normalized value reflects the uncertainty of the internal estimated temperature of the current cycle. The temperature margin ratio is multiplied by the estimated variance normalized value to generate a thermal damage risk factor. When the estimation uncertainty increases, the risk factor increases accordingly, so that the protection decision automatically tends to be conservative when the information is insufficient. A continuous thermal damage risk index sequence is generated by applying sliding window mean filtering to the thermal damage risk factors of multiple consecutive working cycles. Sliding window mean filtering can smooth out the interference of single-cycle random fluctuations on risk quantification, so that the risk index sequence can smoothly reflect the evolution direction of overheating trend.

[0034] Furthermore, in a preferred embodiment of the present invention, the step of intervening with multi-mode cooling based on the overheating risk sequence to generate a continuous protection action plan for execution, and adjusting the protection action plan by performing cyclic risk assessment during execution, specifically includes: The continuous thermal damage risk index sequence is divided into four risk intervals in ascending order: normal interval, warning interval, restricted interval, and emergency interval. A transition overlap zone is set between adjacent intervals, and the boundary threshold of the interval division is dynamically shifted based on the current load mode identification result. Each index value in the continuous thermal damage risk index sequence is mapped to the corresponding risk interval to form an overheating risk sequence that evolves over time. Each element in the overheating risk sequence contains the risk interval to which the current work cycle belongs and the relative position ratio of the risk index within its respective interval. Based on the risk zone attribution and relative position ratio of each working cycle in the overheat risk sequence, multi-mode cooling intervention is executed to generate a continuous protection action plan; if the current element belongs to the normal range, the original engine control parameters are maintained, the injection pulse width is calculated according to the normal power air-fuel ratio, and the ignition advance angle is kept at the original engine calibration value. If the current element belongs to the warning range, an additional injection pulse width is superimposed on the basic injection pulse width. The amplitude of the additional injection pulse width is linearly positively correlated with the relative position ratio of the risk index within the warning range. The ignition advance angle is delayed and the delay amplitude is linearly correlated with the relative position ratio. The maximum allowable speed of the engine is limited to the preset intermediate speed level. If the current element belongs to the restricted range, the throttle opening will be continuously reduced along the preset decay curve, the fuel injection mode will be changed to periodic intermittent injection, and the non-fuel injection working cycle will retain ignition to maintain the crankshaft rotational inertia, and the engine speed will be maintained at the preset idle speed level to drive the cooling fan to operate. If the current element belongs to the emergency zone, the throttle will be completely closed, cutting off the fuel injection signal and ignition signal. After the engine stops, the operator must manually reset it before it can be restarted. During the execution of the protection action plan, each work cycle will re-determine the cycle risk based on the newly generated thermal damage risk index and dynamically adjust the protection action plan parameters.

[0035] It should be noted that traditional single-threshold triggering methods are prone to frequent start-stop of protection actions at the threshold critical point due to small fluctuations in the thermal damage risk index, causing periodic oscillations in engine speed and deteriorating the user experience. Therefore, the continuous thermal damage risk index sequence is divided into four progressive risk intervals and a transition overlap zone is introduced. The transition overlap zone makes interval switching hysteretic; the risk index must cross the overlap zone and enter the adjacent interval to a certain depth before triggering a level change, thus avoiding boundary oscillations. The boundary thresholds for interval division are not fixed values, but dynamically offset based on the current load mode identification results. In heavy-load cutting mode and saw overload mode, the lower boundary thresholds of the warning interval and the limit interval are automatically reduced by the preset offset, so that the protection triggering timing is advanced as the load increases. This is because the engine thermal limit margin is smaller under heavy-load conditions, requiring earlier intervention to ensure safety margin. In the overheating risk sequence generated after mapping each risk index to the corresponding interval, each element not only contains the interval affiliation but also records the relative position ratio of the risk index within its interval. This ratio allows the intensity of subsequent protection actions to be continuously adjusted rather than abruptly changed, providing a quantitative basis for a smooth and gradual control response.

[0036] Within the normal operating range, the engine's original control parameters remain unchanged. The injection pulse width is calculated based on the normal power air-fuel ratio, and the throttle opening fully responds to the operator's throttle command, ensuring that the chainsaw operates at full power without the risk of overheating. When entering the warning range, an additional injection pulse width is superimposed on the basic injection pulse width. The amplitude of the additional pulse width is linearly positively correlated with the relative position. This absorbs in-cylinder heat through the latent heat of excess fuel vaporization and reduces the temperature of the adiabatic flame. At the same time, the ignition advance angle is linearly retarded to further suppress knocking tendency and reduce thermal load. The maximum permissible speed is limited to a preset intermediate speed level to appropriately reduce power output without interrupting the cutting operation. The operator can actively allow the engine to cool down after completing the current cut.

[0037] When the engine enters the restricted zone, it indicates a significant increase in overheating risk. The throttle opening continuously decreases along a preset decay curve, causing the engine to disconnect from power output. The fuel injection mode changes from continuous injection to periodic intermittent injection. The duty cycle of intermittent injection is inversely proportional to the relative position ratio. During non-injection cycles, ignition is maintained to preserve crankshaft rotational inertia. The engine speed is maintained at a preset idle level to drive the cooling fan to continuously remove heat from the cylinder block. During this stage, cooling is accelerated through a dual mechanism of forced convection cooling and reduced combustion heat generation. When the engine enters the emergency zone, the risk of thermal damage approaches the boundary of irreversible failure. The throttle is completely closed, and fuel supply and ignition are simultaneously cut off. The engine quickly shuts down due to inertia. After shutdown, the operator must manually reset the engine before restarting, forcing a manual confirmation of a safe state. During the entire protection action execution, each work cycle re-assesses the cyclic risk based on the newly generated thermal damage risk index and dynamically adjusts the protection action parameters, forming a closed-loop feedback adjustment mechanism. When the risk index decreases, the protection action strength automatically weakens; when the risk index increases, the protection action strength automatically strengthens, ensuring that the protection response is always precisely matched to the degree of overheating danger.

[0038] Figure 2 A flowchart of a method for diagnosing the degradation of an exhaust temperature sensor based on statistical feature analysis of a novel information sequence, provided in an embodiment of the present invention; like Figure 2 As shown, the present invention provides a flowchart of a method for diagnosing the degradation of an exhaust temperature sensor based on statistical feature analysis of a novelty sequence, comprising: The difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, output by the two-node thermal state space model in each working cycle, constitute an information sequence, which serves as the data source for sensor health status monitoring. Sliding window recursive statistics are performed on the information sequence of a continuous preset number of working cycles. The smoothed values ​​of peak temperature information and valley temperature information are calculated using the exponentially weighted moving average algorithm, and the larger of the two absolute values ​​is taken as the mean drift. The larger of the variance of peak temperature information and the variance of valley temperature information within the window is divided by the factory calibration reference variance to obtain the variance abnormality increase ratio. The first-order autoregressive model coefficients of the peak temperature information sequence within the sliding window were fitted using the Yule-Walker equation. The autocorrelation coefficient of the lag step was extracted and substituted into the exponential decay model. The number of lag steps corresponding to the decay of the normalized autocorrelation function to the preset threshold was calculated and converted into crankshaft rotation angle as the autocorrelation decay time constant. The mean drift, variance aberration increase rate, and autocorrelation decay time constant are compared with preset slight degradation thresholds and moderate degradation thresholds, respectively. If at least one indicator exceeds the slight degradation threshold and no indicator reaches the moderate degradation threshold, it is rated as slight degradation. If at least one indicator exceeds the moderate degradation threshold, it is rated as moderate degradation. When the degradation is assessed as slight, the measured signal is maintained to participate in the closed-loop correction. The multiplication coefficient of the estimated variance normalization value used in calculating the thermal damage risk factor is increased by a preset ratio to make the overheat protection decision more conservative. Warning information containing degradation index values ​​and mild maintenance recommendations is generated and pushed out. When the degradation is rated as moderate, a non-disruptive switch from closed-loop calibration to open-loop prediction is initiated. This causes the inverse matrix norm of the measurement noise covariance matrix to decrease smoothly over multiple consecutive working cycles according to an exponential decay function, so that the Kalman gain decays synchronously to zero. After the switch, a pure open-loop prediction is performed by a two-node thermal state-space model. The safe upper limit of cylinder wall temperature is multiplied by a margin factor, and maintenance information including sensor replacement recommendations is generated and pushed out.

[0039] It should be noted that when the sensor is in a healthy state, the innovation sequence exhibits white noise characteristics with zero mean, low variance, and near-uncorrelated time series. However, when the thermocouple junction degrades due to factors such as surface carbon accumulation, oxide layer thickening, or decreased insulation resistance, the statistical characteristics of the innovation sequence will undergo a systematic shift with directional features. This shift can be detected without injecting additional test signals into the sensor. Therefore, the innovation sequence, which is composed of the difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, output by the dual-node thermal state space model in each working cycle, is used as the data source for sensor health status monitoring, realizing non-invasive online monitoring.

[0040] A sliding window recursive statistical analysis is performed on the information sequence of a continuously preset number of working cycles. An exponentially weighted moving average algorithm is used to calculate the smoothed values ​​of the peak and trough temperature information. This algorithm assigns higher weight to recent windows, enabling it to sensitively capture the trend changes in sensor degradation. The larger of the two absolute values ​​is taken as the mean drift, which can effectively identify unidirectional attenuation or zero-point shift in sensor sensitivity. The larger of the variances of the peak and trough temperature information within the window is divided by the factory calibration reference variance to obtain the variance amplification ratio. This ratio increases significantly when the sensor node becomes loose or the contact resistance fluctuates, sensitively reflecting the degree of abnormal amplification of the signal noise level. The first-order autoregressive model coefficients of the peak temperature information sequence within a sliding window are fitted using the Yule-Walker equation. The autocorrelation coefficient of the lag step is extracted and substituted into the exponential decay model. The number of lag steps corresponding to the normalized autocorrelation function decaying to a preset threshold is calculated and converted into crankshaft angle as the autocorrelation decay time constant. This time constant reflects the degree of dynamic hysteresis in the sensor response. When carbon accumulation on the thermocouple junction surface slows down heat conduction, a persistent same-sign bias appears in the information sequence, increasing the autocorrelation decay time constant. The above three degradation evaluation indicators characterize the sensor's performance degradation state from different dimensions, avoiding potential omissions or misjudgments caused by single-indicator evaluation. The three indicators are compared with preset slight degradation thresholds and moderate degradation thresholds, respectively. Slight degradation assessment requires continuous exceedance for a preset number of windows to prevent instantaneous spikes in a single indicator from causing repeated jumps in the assessment results. Moderate degradation assessment takes effect after a single exceedance to ensure a safe response speed.

[0041] When assessed as slight degradation, the measured signal is maintained for closed-loop correction, but the multiplication coefficient of the estimated variance normalization value is increased by a preset ratio. This causes the thermal damage risk factor to automatically increase as sensor information uncertainty increases, making the overheat protection decision more conservative. When assessed as moderate degradation, a seamless switch from closed-loop correction to open-loop prediction is initiated. The inverse matrix norm of the measurement noise covariance matrix decreases smoothly according to an exponential decay function, and the Kalman gain decreases synchronously and smoothly to zero. After the switch, the dual-node thermal state-space model continues to calculate the internal temperature in pure open-loop prediction mode and multiplies the safe upper limit of cylinder wall temperature by a margin coefficient. The entire transition process is without state jumps, and the protection function is not interrupted, solving the problem of immediate loss of protection function after traditional sensor failure. At the same time, the generated predictive maintenance information provides data support for the full life cycle maintenance of the chainsaw.

[0042] Furthermore, the method for graded overheat protection of a gasoline saw based on exhaust temperature feedback provided by this invention also includes the following steps: The equivalent thermal resistance of the deposited layer is extended to the two-node thermal state space model as the third state variable, and a third-order augmented state model is established. The equivalent thermal resistance of the deposited layer is modeled as a random walk process. The process noise variance is adaptively adjusted according to the concentration level of unburned hydrocarbons in the exhaust gas. When the concentration of unburned hydrocarbons increases, the process noise variance increases accordingly. An augmented measurement equation with correction for equivalent thermal resistance of the deposit layer is established. An augmented state estimator is used to jointly correct the cylinder wall high-temperature node temperature, exhaust port wall node temperature and equivalent thermal resistance of the deposit layer based on the observation information in each working cycle. The a posteriori estimate of the cylinder wall critical point temperature and the identification value of the equivalent thermal resistance of the deposit layer are output. Linear least squares fitting is performed on the sequence of equivalent thermal resistance identification values ​​of the deposition layer within the preset evaluation period to extract the deposition layer growth rate. When the equivalent thermal resistance identification value of the deposition layer exceeds the preset cleaning reminder threshold, a cleaning reminder message is generated. After the cleaning operation is detected to be completed, the equivalent thermal resistance of the deposition layer is reset to zero. When the equivalent thermal resistance identification value of the deposited layer exceeds the preset protection intervention threshold and no cleaning operation is performed, the equivalent thermal resistance identification value of the deposited layer is substituted into the inverse transfer function to reconstruct the measured peak temperature and measured valley temperature in reverse. The reconstructed temperature value replaces the original measured value to participate in the closed-loop correction. While initiating reverse reconstruction, the safe upper limit of cylinder wall temperature is multiplied by a margin coefficient that monotonically decreases as the equivalent thermal resistance of the deposit layer increases. The margin coefficient is set to one when the equivalent thermal resistance of the deposit layer is equal to the protection intervention threshold. After the cleaning operation is completed, the safe upper limit of cylinder wall temperature is restored to the standard value when the equivalent thermal resistance of the deposit layer is reset to zero.

[0043] It should be noted that during prolonged cutting operations on high-resin-content wood, a deposit layer gradually accumulates on the surface of the thermocouple at the exhaust port of a gasoline saw. This deposit layer is composed of carbides formed by the high-temperature decomposition of heavy components in the unburned mixture and resinous colloids. The thermal conductivity of this deposit layer is much lower than that of the thermocouple's metal sheath material, forming a dynamically growing thermal resistance barrier between the thermocouple junction and the exhaust gas flow. The direct consequence of this thermal resistance barrier is that the thermocouple's response to the peak temperature of the exhaust pulse exhibits amplitude attenuation, and its response to the valley temperature after scavenging shows baseline elevation. Furthermore, the response delay to temperature changes continues to increase as the deposit layer thickens. However, the sensor itself is not completely ineffective; its output signal still changes with temperature, only the measurement accuracy and dynamic characteristics continue to deteriorate.

[0044] Therefore, based on the aforementioned two-node thermal state space model, this scheme equates the thermal resistance effect of the deposition layer to a first-order heat conduction link connected in series between the thermocouple junction and the exhaust gas flow. The equivalent thermal resistance monotonically increases with the deposition layer thickness. This equivalent thermal resistance is then augmented as a third state variable into the original state vector to establish a third-order augmented state model. The state transition equation for the equivalent thermal resistance of the deposition layer is modeled as a random walk process. The process noise variance is adaptively adjusted based on the concentration level of unburned hydrocarbons in the exhaust gas. When the throttle opening increases, leading to a richer mixture, or when drastic speed fluctuations exacerbate scavenging short-circuiting, the amount of unburned hydrocarbons emitted increases, and the process noise variance increases accordingly to match the accelerated growth pattern of the deposition layer. The augmented measurement equation connects a heat conduction attenuation factor determined by the equivalent thermal resistance in series with the original first-order inertial link, ensuring that the predicted peak and valley temperatures reflect the amplitude attenuation and phase lag caused by the thickening of the deposition layer. The augmented state estimator uses observed information to jointly correct three state variables in each work cycle, simultaneously outputting the posterior estimate of the cylinder wall critical temperature and the identification value of the equivalent thermal resistance of the deposit layer. Linear least-squares fitting is performed on the equivalent thermal resistance identification value sequence within a preset evaluation period to extract the deposit growth rate. When the equivalent thermal resistance exceeds the cleaning reminder threshold, a cleaning reminder message is generated and pushed to the operator's terminal. After cleaning, the exhaust temperature signal characteristics are automatically detected and the equivalent thermal resistance is reset to zero. If the equivalent thermal resistance exceeds the protection intervention threshold and the operator has not yet cleaned, the identification value is substituted into the inverse transfer function to reconstruct the measured temperature. The compensated reconstructed temperature value replaces the original measured value in the closed-loop correction. Simultaneously, the safe upper limit of cylinder wall temperature is multiplied by a margin coefficient that monotonically decreases with increasing equivalent thermal resistance. When the equivalent thermal resistance equals the protection intervention threshold, the margin coefficient is one. When cleaning is completed and the equivalent thermal resistance is reset to zero, the standard safe upper limit is restored. This forms a closed-loop protection chain from sedimentation layer condition identification, cleaning reminders, reverse compensation to safety limit linkage, ensuring the reliability of overheat protection during the transition period when the sensor is partially unusable due to carbon buildup.

[0045] Furthermore, the method for graded overheat protection of a gasoline saw based on exhaust temperature feedback provided by this invention also includes the following steps: The pulse interval sequence and throttle opening signal of the crankshaft position sensor are collected. The instantaneous crankshaft speed sequence is subjected to a sliding window short-time Fourier transform and the low-frequency power spectral density integral value is extracted as the feed pressure fluctuation index. The number of times the throttle opening falls from the high opening range to the low opening range within a preset time period is counted as the oil return frequency. The peak torque speed range is determined based on the engine's external characteristic curve. The ratio of the duration of the instantaneous speed being continuously lower than the lower limit of the peak torque speed range within a single cutting stroke to the total duration of the cutting stroke is used as the percentage of peak torque duration. The start and end points of the cutting stroke are defined by the moment when the throttle opening crosses the cutting opening threshold and falls back to the idle speed opening. The behavior feature vector is composed of the saw pressure fluctuation index, oil return frequency, and peak torque duration ratio. It is input into a behavior pattern classifier based on fuzzy reasoning. The classifier defines three types of behavior patterns: stable, aggressive, and unfamiliar. The output is the behavior pattern classification result and confidence level. When the classification result is either radical or unfamiliar and the confidence level exceeds the preset confidence threshold, the first difference of the thermal damage risk index is calculated and the current rate of change is obtained by smoothing it with an exponentially weighted moving average. The estimated remaining time is obtained by dividing the difference between the current value of the thermal damage risk index and the lower limit of the warning interval by the current rate of change. When the estimated remaining time is less than the preset guidance trigger time threshold, the built-in vibration motor of the drive handle emits a vibration reminder signal. The vibration frequency is inversely proportional to the estimated remaining time. At the same time, it pushes operation behavior improvement suggestions corresponding to the behavior mode category to the display terminal. If the saw pressure fluctuation index decreases or the return oil frequency increases to the preset improvement level within the preset response time, it is determined that the operator has responded and the intervention is cancelled; otherwise, graded protection is implemented when the thermal damage risk index enters the warning range.

[0046] It should be noted that in actual operation of chainsaws, operator behavior has a significant impact on engine thermal status. Novice operators often apply excessive feed pressure, leading to a sharp increase in frictional heat. Traditional overheat protection only passively responds to temperature thresholds, failing to incorporate operator behavior into the decision-making process, thus missing the opportunity to guide behavioral changes and prevent protection triggering at the initial stage of thermal risk. Therefore, this solution establishes an operator behavior characteristic identification and overheat prevention guidance mechanism, extracting three behavioral feature quantities from the crankshaft position sensor pulse interval sequence and throttle opening signal. A sliding window short-time Fourier transform is performed on the instantaneous crankshaft speed sequence, and the low-frequency power spectral density integral value is extracted as the feed pressure fluctuation index. Changes in the operator's arm force modulate the sawing resistance, thereby causing changes in the power distribution of the crankshaft's instantaneous speed in the low-frequency range. The higher and more uneven the feed pressure, the higher this integral value. The oil return frequency is obtained by counting the number of times the throttle opening drops from the high opening range to the low opening range. A transition zone is set between the two ranges to avoid miscounting caused by slight hand tremors. A low oil return frequency indicates that the operator is continuously applying pressure without actively retracting the blade. The peak torque speed range is determined based on the engine's external characteristic curve. The percentage of time the instantaneous speed within a single cutting stroke remains below the lower limit of this range is counted as the peak torque duration percentage. The start and end points of the cutting stroke are defined by the moment when the throttle opening crosses the cutting opening threshold and drops back to the idle speed. A higher percentage indicates that the operator is operating in the dangerous low-speed, high-torque range for an extended period.

[0047] Three features are input into a behavior pattern classifier based on fuzzy inference. The classifier defines three patterns: stationary, aggressive, and unfamiliar. The membership function parameters are calibrated offline by collecting behavioral data from operators with different experience levels. The classification results and confidence scores are output. When the classification is aggressive or unfamiliar and the confidence score exceeds the threshold, the first difference of the thermal damage risk index is calculated and smoothed to obtain the rate of change. The difference between the current value and the lower limit of the warning interval is divided by the rate of change to obtain the estimated remaining time. This quantifies how long the operator can continue working according to the current operating habits before triggering protection. When the estimated remaining time is less than the guidance trigger threshold, the built-in vibration motor of the drive handle vibrates at a frequency inversely proportional to the estimated remaining time to remind the operator. At the same time, improvement suggestions corresponding to the behavior pattern are pushed, forming a dual guidance of tactile and visual senses. If the saw pressure fluctuation index decreases or the return oil frequency increases to the required improvement level within the preset response time, the intervention is canceled. Otherwise, graded protection is implemented when the risk index enters the warning interval. The above approach extends overheat protection from a passive response to an active prevention approach. By guiding operators to reduce heat load at the source during the risk incubation stage, it reduces the frequency of protection triggering and improves continuous operation efficiency.

[0048] Figure 3 An embodiment of the present invention provides a gasoline saw overheating graded protection system 3 based on exhaust temperature feedback. The system includes: a memory 301, a processor 302, and a communication interface 303. The memory 301 contains a gasoline saw overheating graded protection method program based on exhaust temperature feedback. When the gasoline saw overheating graded protection method program based on exhaust temperature feedback is executed by the processor 302, it implements the gasoline saw overheating graded protection method steps as described in any of the above claims.

[0049] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for graded overheat protection of a gasoline saw based on exhaust temperature feedback, characterized in that, include: Collect fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals. Divide the window according to the working cycle and perform dynamic segmentation of the exhaust temperature waveform within the cycle. Simultaneously extract the working cycle feature parameters to construct a multi-dimensional feature spectrum vector. A two-node thermal state space model is constructed. The multidimensional feature spectrum vector of each working cycle is used as the sequential observation input to estimate the time-varying parameters of thermal state transfer between cycles. The prior state is corrected by calculating the observation information and the internal temperature is reconstructed to obtain the internal predicted temperature. The instantaneous crankshaft speed sequence is obtained, and multi-parameter phase space mapping of load modes is performed by combining multi-dimensional feature spectrum vectors. Load mode recognition is performed in the mapping space using clustering algorithms. Based on real-time clustering results and internal predicted temperature, thermal damage risk factors are constructed to classify and quantify thermal damage risk, and a continuous thermal damage risk index sequence is generated. Based on the preset overheating risk judgment rules, the continuous thermal damage risk index sequence is divided into different risk intervals to obtain the overheating risk sequence that evolves over time. Based on the overheating risk sequence, multi-mode cooling intervention is performed to generate a continuous protection action plan for execution. During the execution, cyclic risk judgment is performed to adjust the protection action plan. During the continuous operation of the overheat protection, the long-term health status of the exhaust temperature sensor is monitored non-invasively. The degree of degradation is determined by the mean drift, the percentage increase in variance, and the autocorrelation decay time constant. Degradation correction is performed based on the degree of degradation, and predictive maintenance information is generated and pushed out.

2. The method for graded overheat protection of a gasoline saw based on exhaust temperature feedback according to claim 1, characterized in that, The method involves acquiring fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals, dividing the window according to the working cycle and performing dynamic segmentation of the exhaust temperature waveform within the cycle, and simultaneously extracting working cycle feature parameters to construct a multi-dimensional feature spectrum vector, specifically including: Collect fast-response thermocouple signals from the exhaust port of the target gasoline saw and crankshaft phase pulse signals. Input the fast-response thermocouple signals to an analog-to-digital converter to convert them into digital voltage sequences at a preset sampling rate. Based on the thermocouple indexing characteristics, map the digital voltage sequences into exhaust temperature sampling sequences. Synchronously convert the crankshaft phase pulse signal into an edge-triggered sequence and record the pulse interval between adjacent top dead center points. Based on the pulse count, divide the exhaust temperature sampling sequence into several independent working cycle windows according to the preset crankshaft rotation angle. For the exhaust temperature sampling sequence within each working cycle window, the local standard deviation of each sampling point is calculated by sliding the fixed narrow window width point by point as the instantaneous noise base estimate. The noise base estimate is then arranged along the time axis to form the noise base curve within the current cycle. Based on the noise floor curve, an exhaust pulse determination rule is established, the first-order forward difference of the exhaust temperature sampling sequence is calculated, and the exhaust pulse rise start point, exhaust pulse peak point and exhaust pulse end point are marked using the exhaust pulse determination rule. Starting from the end point of the exhaust pulse, continue tracking to the lowest temperature region where the exhaust temperature fluctuation is less than the first threshold and the first-order difference fluctuates narrowly near zero. Search for the minimum sample value in the lowest temperature region as the valley temperature point after scavenging, and complete the calibration of the exhaust pulse peak point, valley temperature point, rising start point and end point within the current working cycle window. Based on the calibration results, the peak temperature, valley temperature, maximum gradient of rising and falling edges, peak half width, pulse area integral, and asymmetry ratio are extracted and combined in sequence to form the feature vector sequence of the current working cycle. After performing directional adaptive filtering on the feature vector sequences of several consecutive working cycles, a multidimensional feature spectrum vector is obtained.

3. The method for graded overheat protection of a gasoline saw based on exhaust temperature feedback according to claim 1, characterized in that, The time-varying parameter estimation for inter-cycle thermal state transfer involves calculating observational information to correct prior states and reconstructing the internal temperature to obtain the predicted internal temperature. Specifically, this includes: A two-node thermal state space model is constructed with cylinder wall high-temperature nodes and exhaust port wall nodes as state variables. The cylinder wall high-temperature nodes correspond to the cylinder wall region near the exhaust port, and the exhaust port wall nodes correspond to the inner wall region of the exhaust port. Each node has independent heat capacity and temperature state. The nodes are coupled through heat conduction terms and have convective heat transfer paths with the external environment. Taking the first working cycle after the cold start of the gasoline saw as the starting point, the initial temperatures of the high-temperature nodes of the cylinder wall and the exhaust port wall nodes are set to the measured values ​​of the ambient temperature sensor. The initial state covariance matrix is ​​set as a diagonal matrix, and the values ​​of the diagonal elements represent the degree of uncertainty of the initial temperature estimation. Before each state prediction, the convective heat transfer coefficient in the state transition matrix is ​​updated in real time based on the engine speed and throttle opening of the current working cycle. The multidimensional feature spectrum vector of the current working cycle is used as the sequential observation input, and the peak temperature and valley temperature are used as the dual measurement outputs to establish the peak temperature measurement equation and the valley temperature measurement equation. After completing the state transition matrix update and measurement equation establishment, the prior state prediction is performed. The posterior temperature estimates of the cylinder wall high temperature node and exhaust port wall node output from the previous working cycle are used as the initial state of this cycle. They are substituted into the updated state transition matrix to obtain the prior temperature estimate and the prior state covariance matrix. Substitute the prior temperature estimate into the peak temperature measurement equation and the valley temperature measurement equation to calculate the predicted peak temperature and the predicted valley temperature. Calculate the difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, respectively, to form the observation information vector. The deviation amplitude between the peak temperature and the valley temperature in the multidimensional feature spectrum vector before and after directional adaptive filtering is calculated. The measurement noise covariance matrix is ​​adjusted according to the deviation amplitude. The Kalman gain is calculated using the adjusted measurement noise covariance. The prior temperature estimate is corrected by weighting the observation information vector with the Kalman gain to obtain the posterior temperature estimate. The posterior temperature estimate of the high temperature node of the cylinder wall is taken as the internal predicted temperature.

4. The method for graded overheat protection of a gasoline saw based on exhaust temperature feedback according to claim 1, characterized in that, The process involves multi-parameter phase space mapping of load modes, using clustering algorithms to identify load modes within the mapping space, and constructing thermal damage risk factors based on real-time clustering results and internally predicted temperatures for thermal damage risk grading and quantification. Specifically, this includes: Extract the instantaneous crankshaft speed sequence of each working cycle from the pulse interval sequence of the crankshaft position sensor, which shows the instantaneous speed change with the crankshaft angle. Calculate the difference between the highest and lowest instantaneous speeds within the cycle to obtain the peak-to-peak speed. Perform a Fourier transform on the instantaneous crankshaft speed sequence and calculate the proportion of half-order harmonic energy. Extract the crankshaft angle phase at the moment when the exhaust peak temperature occurs and the crankshaft angle phase at the moment when the speed peak occurs, and calculate the difference between the two as the phase offset. The moment when the exhaust peak temperature occurs is determined by the calibration result of the exhaust pulse peak point corresponding to the multidimensional feature spectrum vector. The peak half-width and the maximum gradient at the falling edge of the current working cycle are extracted from the multidimensional feature spectrum vector. The peak half-width and the maximum gradient at the falling edge are multiplied to obtain the single-cycle exhaust thermal impulse characterization value. This value, together with the peak-to-peak speed, the proportion of half-order harmonic energy, and the phase offset, constitutes a four-dimensional load feature vector, which serves as the real-time mapping point of the current working cycle in the multi-parameter phase space. Four cluster centers—empty mode, light-load cutting mode, heavy-load cutting mode, and cam saw overload mode—are preset as initial benchmarks. The Euclidean distance between the current four-dimensional load feature vector and the four preset cluster centers is calculated using the nearest neighbor classification algorithm. The load mode corresponding to the preset cluster center with the smallest Euclidean distance is selected as the classification result of the current working cycle, thus obtaining the load mode identification information. Based on the load mode recognition information, the upper limit value of cylinder wall temperature corresponding to the current load mode is retrieved from the preset safety limit parameter group, and the internal estimated temperature is divided by the upper limit value of cylinder wall temperature to obtain the temperature margin ratio. Meanwhile, the diagonal elements of the posterior variance corresponding to the high-temperature nodes of the cylinder wall are extracted from the posterior state covariance matrix of the two-node thermal state space model. The estimated normalized variance value is obtained by normalizing the diagonal elements corresponding to the initial covariance matrix of the cold start as a benchmark. The thermal damage risk factor for the current work cycle is obtained by multiplying the temperature margin ratio by the estimated variance normalized value. The thermal damage risk factors for multiple consecutive work cycles are then subjected to sliding window mean filtering to generate a continuous thermal damage risk index sequence.

5. The method for graded overheat protection of a gasoline saw based on exhaust temperature feedback according to claim 1, characterized in that, The process involves multi-mode cooling intervention based on overheating risk sequences to generate continuous protection action plans for execution. During execution, cyclical risk assessments are performed to adjust the protection action plans. Specifically, this includes: The continuous thermal damage risk index sequence is divided into four risk intervals in ascending order: normal interval, warning interval, restricted interval, and emergency interval. A transition overlap zone is set between adjacent intervals, and the boundary threshold of the interval division is dynamically shifted based on the current load mode identification result. Each index value in the continuous thermal damage risk index sequence is mapped to the corresponding risk interval to form an overheating risk sequence that evolves over time. Each element in the overheating risk sequence contains the risk interval to which the current work cycle belongs and the relative position ratio of the risk index within its respective interval. Based on the risk zone attribution and relative position ratio of each working cycle in the overheat risk sequence, multi-mode cooling intervention is executed to generate a continuous protection action plan; if the current element belongs to the normal range, the original engine control parameters are maintained, the injection pulse width is calculated according to the normal power air-fuel ratio, and the ignition advance angle is kept at the original engine calibration value. If the current element belongs to the warning range, an additional injection pulse width is superimposed on the basic injection pulse width. The amplitude of the additional injection pulse width is linearly positively correlated with the relative position ratio of the risk index within the warning range. The ignition advance angle is delayed and the delay amplitude is linearly correlated with the relative position ratio. The maximum allowable speed of the engine is limited to the preset intermediate speed level. If the current element belongs to the restricted range, the throttle opening will be continuously reduced along the preset decay curve, the fuel injection mode will be changed to periodic intermittent injection, and the non-fuel injection working cycle will retain ignition to maintain the crankshaft rotational inertia, and the engine speed will be maintained at the preset idle speed level to drive the cooling fan to operate. If the current element belongs to the emergency zone, the throttle will be completely closed, cutting off the fuel injection signal and ignition signal. After the engine stops, the operator must manually reset it before it can be restarted. During the execution of the protection action plan, each work cycle will re-determine the cycle risk based on the newly generated thermal damage risk index and dynamically adjust the protection action plan parameters.

6. The method for graded overheat protection of a gasoline saw based on exhaust temperature feedback according to claim 1, characterized in that, The non-invasive monitoring of the long-term health status of the exhaust temperature sensor, using mean drift, variance aberration increase rate, and autocorrelation decay time constant to determine the degree of degradation, performs degradation correction based on the degree of degradation, and generates predictive maintenance information for push notification, specifically including: The difference between the measured peak temperature and the predicted peak temperature, and the difference between the measured valley temperature and the predicted valley temperature, output by the two-node thermal state space model in each working cycle, constitute an information sequence, which serves as the data source for sensor health status monitoring. Sliding window recursive statistics are performed on the information sequence of a continuous preset number of working cycles. The smoothed values ​​of peak temperature information and valley temperature information are calculated using the exponentially weighted moving average algorithm, and the larger of the two absolute values ​​is taken as the mean drift. The larger of the variance of peak temperature information and the variance of valley temperature information within the window is divided by the factory calibration reference variance to obtain the variance abnormality increase ratio. The first-order autoregressive model coefficients of the peak temperature information sequence within the sliding window were fitted using the Yule-Walker equation. The autocorrelation coefficient of the lag step was extracted and substituted into the exponential decay model. The number of lag steps corresponding to the decay of the normalized autocorrelation function to the preset threshold was calculated and converted into crankshaft rotation angle as the autocorrelation decay time constant. The mean drift, variance aberration increase rate, and autocorrelation decay time constant are compared with preset slight degradation thresholds and moderate degradation thresholds, respectively. If at least one indicator exceeds the slight degradation threshold and no indicator reaches the moderate degradation threshold, it is rated as slight degradation. If at least one indicator exceeds the moderate degradation threshold, it is rated as moderate degradation. When the degradation is assessed as slight, the measured signal is maintained to participate in the closed-loop correction. The multiplication coefficient of the estimated variance normalization value used in calculating the thermal damage risk factor is increased by a preset ratio to make the overheat protection decision more conservative. Warning information containing degradation index values ​​and mild maintenance recommendations is generated and pushed out. When the degradation is rated as moderate, a non-disruptive switch from closed-loop calibration to open-loop prediction is initiated. This causes the inverse matrix norm of the measurement noise covariance matrix to decrease smoothly over multiple consecutive working cycles according to an exponential decay function, so that the Kalman gain decays synchronously to zero. After the switch, a pure open-loop prediction is performed by a two-node thermal state-space model. The safe upper limit of cylinder wall temperature is multiplied by a margin factor, and maintenance information including sensor replacement recommendations is generated and pushed out.

7. A graded overheat protection system for a gasoline saw based on exhaust temperature feedback, characterized in that, The system includes: a memory, a processor, and a communication interface. The memory contains a program for a gasoline saw overheating graded protection method based on exhaust temperature feedback. When the program for the gasoline saw overheating graded protection method based on exhaust temperature feedback is executed by the processor, it implements the steps of the gasoline saw overheating graded protection method based on exhaust temperature feedback as described in any one of claims 1-6.

Images