A method for identifying the wear state of a cutter of a saline-alkali soil straw field machine

Abstract

The application discloses a kind of salt alkali soil straw field machine tool wear state identification method, it is related to agricultural machinery intelligent monitoring technical field.The method comprises the following steps: obtaining surface soil conductivity value and moisture content value;Actual torque, tool shaft actual speed, tool shaft bearing seat vibration acceleration and travel speed are collected during operation;The ratio of actual torque and actual speed is calculated, and the equivalent load characteristic value is obtained;Fourier transform is carried out on vibration acceleration signal, and the amplitude of fundamental frequency and double frequency is extracted as vibration characteristic value;Equivalent load characteristic value, vibration characteristic value, soil conductivity, soil moisture content, travel speed, load-vibration coupling coefficient and salt-velocity erosion index are input into neural network model, and wear probability value is output;When the probability value is greater than the preset threshold, it is determined that the tool needs to be replaced.The application improves the identification accuracy and operation continuity by fusing multi-source mechanics, dynamics and soil environment parameters.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for agricultural machinery, specifically relating to a method for identifying the wear status of blades in a straw returning machine for saline-alkali land. Background Technology

[0002] Returning straw to the field is one of the core technologies of conservation tillage, playing a vital role in increasing soil organic matter content and improving soil structure. As the working parts that directly contact the soil and straw, the blades of straw returning machines are subjected to multiple effects during operation, including abrasive wear, impact loads, and corrosion. Their wear condition directly affects the straw crushing quality, operating efficiency, and equipment lifespan. Therefore, accurate and timely identification of the wear condition of straw returning machine blades is crucial for ensuring operational quality and reducing maintenance costs.

[0003] Currently, in agricultural production, the assessment of blade wear on straw returning machines primarily relies on the operator's experience. This includes listening to abnormal sounds from the cutter shaft, visually assessing the uniformity of straw crushing after operation, or periodically checking the remaining thickness of the blade edges after shutdown. Some models attempt to use power output shaft current monitoring to indirectly determine blade load through current changes. However, manual judgment is highly subjective, with significant differences in standards. Shutdown inspections are time-consuming, and the current signal is easily affected by factors such as engine speed fluctuations and transmission system efficiency degradation, making it difficult to accurately characterize the actual wear process of the blades. While existing technologies employ vibration signal monitoring or machine learning methods for blade wear identification, they largely focus on analyzing single mechanical signals and fail to fully consider the impact of soil environmental factors on blade wear, making them unsuitable for complex operating conditions such as saline-alkali land.

[0004] Therefore, there is an urgent need for a tool wear state identification method that can integrate multi-source mechanical signals and saline-alkali soil environmental parameters to overcome the above-mentioned problems in the existing technology. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings in the existing technology by providing a method for identifying the wear state of the cutter blades of a straw returning machine in saline-alkali land. The method integrates multi-source mechanical signals such as power output shaft torque, cutter shaft speed, vibration acceleration, and travel speed, and combines them with environmental parameters of saline-alkali land such as surface soil conductivity and moisture content. This enables accurate and real-time identification of the cutter wear state under complex working conditions in saline-alkali land, effectively avoiding misjudgments and omissions caused by sudden changes in soil environment or interference from a single signal, and significantly improving identification accuracy and operational continuity.

[0006] To solve the above problems, the present invention adopts the following technical solution: A method for identifying the wear condition of blades in a straw returning machine for saline-alkali land includes the following steps: S1, obtain the surface soil electrical conductivity and surface soil moisture content of the current working site; S2, during the operation of the straw returning machine, collect the actual torque value of the power output shaft, the actual rotation speed value of the cutter shaft, the vibration acceleration signal at the cutter shaft bearing seat, and the travel speed value of the straw returning machine; S3, calculate the equivalent load characteristic value based on the actual torque value and the actual speed value; S4, perform frequency domain transformation on the vibration acceleration signal, and extract the amplitude corresponding to the fundamental frequency and the second harmonic amplitude of the tool shaft rotation frequency as vibration characteristic values; S5, the equivalent load characteristic value, the vibration characteristic value, the surface soil electrical conductivity value, the surface soil moisture content value, the travel speed value, the load-vibration coupling coefficient, and the salt-velocity erosion index are input into a pre-trained neural network model, and the wear degree probability value is output; wherein, the load-vibration coupling coefficient is constructed based on the equivalent load characteristic value and the vibration characteristic value, and the salt-velocity erosion index is constructed based on the surface soil electrical conductivity value and the travel speed value; S6. When the wear probability value is greater than a preset threshold, the tool wear condition is determined to be in a state that needs to be replaced.

[0007] Further, in step S1, multiple sampling points are selected within the current work plot in a grid or diagonal pattern. The surface soil electrical conductivity and surface soil moisture content of each sampling point are measured, and the location coordinates of each sampling point are recorded. The location coordinates of each sampling point and its corresponding electrical conductivity and moisture content are stored in a plot parameter lookup table. During the operation of the straw returning machine, the current work position is obtained through a positioning device. Based on the spatial distance between the current work position and the locations of each sampling point in the plot parameter lookup table, the electrical conductivity and moisture content corresponding to the nearest sampling point are selected as the surface soil electrical conductivity and surface soil moisture content.

[0008] Furthermore, in step S2, the acquisition of the actual torque value, the actual rotational speed value, the vibration acceleration signal, and the travel speed value are all carried out during the stable operating period of the straw returning machine; the stable operating period is determined by the real-time change rate of the travel speed value being less than a preset speed fluctuation threshold and the real-time change rate of the actual rotational speed value being less than a preset rotational speed fluctuation threshold.

[0009] Further, in step S3, multiple sets of actual torque values ​​and actual speed values ​​are continuously collected within a preset time window, and the instantaneous ratio of each set of actual torque values ​​to the corresponding actual speed values ​​is calculated. The median of multiple instantaneous ratios is taken as the equivalent load characteristic value.

[0010] Further, in step S4, at least three complete cycles of sub-signals are continuously extracted from the vibration acceleration signal with the reciprocal of the rotation frequency of the cutter shaft as the period. The extracted sub-signals are then synchronously averaged in the time domain to obtain the averaged single-cycle vibration waveform. The averaged single-cycle vibration waveform is then subjected to the frequency domain transformation to extract the amplitude corresponding to the fundamental frequency and the amplitude of the second harmonic, which are used as the vibration characteristic values.

[0011] Furthermore, the neural network model includes: The feature input layer is used to normalize the input equivalent load feature value, vibration feature value, surface soil electrical conductivity value, surface soil moisture content value and travel speed value respectively, and generate the load-vibration coupling coefficient and the salt-velocity erosion index based on the normalized values. The first branch is used to input the normalized equivalent load characteristic value, the vibration characteristic value, the surface soil electrical conductivity value, the surface soil moisture content value, and the travel speed value, and output the first wear probability value. The second branch is used to input the load-vibration coupling coefficient and the salt-velocity erosion index, and output the second wear probability value. The third branch is used to input the normalized surface soil electrical conductivity value and the surface soil moisture content value, and output the third wear probability value. The fusion output layer is used to fuse and output the final wear probability value based on the first wear probability value, the second wear probability value, and the third wear probability value.

[0012] Further, the fusion output layer calculates the range of the first wear probability value, the second wear probability value, and the third wear probability value; when the range is less than a preset dispersion threshold, the arithmetic mean of the first wear probability value, the second wear probability value, and the third wear probability value is used as the wear degree probability value; when the range is greater than or equal to the preset dispersion threshold, the median of the first wear probability value, the second wear probability value, and the third wear probability value is used as the wear degree probability value.

[0013] Furthermore, during each operation, based on the actual torque value and the actual rotational speed value, the instantaneous cutting power is calculated by dividing the instantaneous power by the product of the torque and the rotational speed by a preset coefficient. The instantaneous cutting power is integrated over time to obtain the total cumulative cutting power from the initial use of the tool to the current time. The total cumulative cutting power is input into a pre-calibrated threshold decay function to obtain the preset threshold at the current time. The threshold decay function is an exponential decay function.

[0014] Further, in step S6, a sliding time window is set, which contains the wear probability value corresponding to L consecutive sampling times; when the wear probability value of at least K sampling times in the sliding time window is greater than the preset threshold, the tool wear state is determined to be a state that needs to be replaced; where L is the total number of sampling times in the window, K is the preset trigger threshold, and K≤L.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This invention incorporates the unique effects of soil salinity and moisture content in saline-alkali soils on tool wear into a discrimination model by acquiring surface soil conductivity and moisture values. This overcomes the high misjudgment rate caused by traditional methods that rely solely on mechanical signals and ignore environmental factors, significantly improving the accuracy and environmental adaptability in saline-alkali soil conditions. Furthermore, by simultaneously acquiring torque, rotational speed, and vibration acceleration, equivalent load and vibration characteristics are constructed, along with a load-vibration coupling feature. This allows for the simultaneous characterization of increased tool cutting resistance and enhanced periodic impact on the tool shaft, improving sensitivity to tool wear, local defects, and uneven wear, and avoiding the vulnerability of single signals to interference.

[0016] This invention constructs a salinity-velocity erosion index by cross-referencing surface soil electrical conductivity and travel speed. This allows the model to quantify the accelerated erosion effect of the combined effects of high salinity and high-speed operation on cutting tools under saline-alkali soil conditions. It directly reflects the nonlinear interaction between load, vibration, salinity, and velocity, improving the model's physical interpretability and adaptability to different working conditions. The neural network model employs a multi-branch structure, learning wear patterns from three levels: all features, cross-features, and environmental features. An adaptive discrete fusion layer is used; when the outputs of the three branches are consistent, the average value is used to retain all information; when there is significant divergence, the median is used to suppress outlier interference, significantly improving the robustness and reliability of the final probability value.

[0017] This invention monitors the rate of change of travel speed and the rate of change of tool spindle speed in real time, allowing data acquisition only when both are in a stable state. This effectively eliminates abnormal data under unstable working conditions such as starting, stopping, turning, stalling, and sudden changes in impact load, improving the reliability of the input data for the identification model. Simultaneously, by integrating instantaneous cutting power over time to obtain cumulative cutting work, and using an exponential decay function to adaptively reduce the alarm threshold with the actual workload of the tool, it avoids premature or late alarms caused by fixed thresholds. Furthermore, by incorporating a majority voting mechanism within a sliding time window, it requires at least K instances of exceeding the threshold in multiple consecutive samples before triggering a replacement decision, effectively filtering out occasional threshold-exceeding events, significantly reducing the false alarm rate, and improving the reliability and credibility of the judgment results. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method for identifying the wear status of the blades in a straw returning machine for saline-alkali land according to the present invention; Figure 2 This is a framework diagram of the neural network model of the present invention, which shows the structure of the feature input layer, the first branch, the second branch, the third branch and the fusion output layer; Figure 3 This is a graph showing the change of the dynamic threshold of the present invention with cumulative cutting work. Figure 4 This is a schematic diagram of the sliding time window used in this invention to determine the tool change status. Detailed Implementation

[0019] The technical solutions of the invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the invention, not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without creative effort are within the scope of the invention.

[0020] In the description of this invention, it should be noted that the terms "above" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience and simplification of the description and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0021] In the description of this invention, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0022] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connection," "setting," "installation," "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0023] like Figure 1 As shown, this invention discloses a method for identifying the wear condition of blades in a straw returning machine for saline-alkali land, comprising the following steps: S1: Obtaining soil parameters: Before the straw returning machine begins operation, a portable soil conductivity meter and soil moisture sensor are used to measure the soil conductivity and moisture content of the surface layer of the current working plot. The measurement depth is 0–10 cm in the surface soil. The two measured values ​​are then input into the register of the on-board controller as environmental parameters.

[0024] In one embodiment of the present invention, the electrical conductivity and moisture content values ​​of the surface soil are obtained as follows: Multiple sampling points are selected within the current work area using either a grid pattern or a diagonal arrangement. When arranged in a grid pattern, points can be evenly distributed within the area at a grid spacing of 20 m × 20 m. When arranged diagonally, a sampling point can be placed every 15 m along the two diagonals of the area. At each sampling point, a portable conductivity meter and a moisture meter are used to measure the conductivity and moisture content at a depth of 0–10 cm in the surface layer. Simultaneously, real-time dynamic differential GPS (RTK-GPS) is used to record the latitude and longitude coordinates of each sampling point.

[0025] The location coordinates of each sampling point, along with their corresponding electrical conductivity and moisture content values, are stored in a lookup table for land parameters. This lookup table is stored in the non-volatile memory of the onboard controller. During the actual operation of the straw returning machine, the current operating position coordinates are acquired in real time via a positioning antenna mounted on the roof and a differential GPS receiver. The controller reads the current position every preset time interval (e.g., 0.5 seconds) and then calculates the Euclidean spatial distance between this position and the coordinates of each sampling point in the lookup table. All calculated distance values ​​are compared, and the sampling point with the smallest distance is selected. The electrical conductivity and moisture content values ​​stored at this sampling point are then retrieved and used as the soil parameters input into the neural network model at the current moment.

[0026] The aforementioned nearest neighbor matching method can quickly obtain representative soil parameters of the current working location without interrupting the operation, overcome the errors caused by spatial variability of soil salinity and moisture, and improve the spatial adaptability and representativeness of soil input parameters.

[0027] S2: Acquisition of multi-source sensor signals: During the operation of the straw returning machine, a torque sensor mounted on the power output shaft continuously collects the actual torque value. The torque sensor can be a strain gauge type with a range of 0–2000 N·m. A speed sensor mounted on the end of the cutter shaft continuously collects the actual rotational speed of the cutter shaft. This speed sensor can be a Hall effect sensor or a magnetoelectric sensor with a measurement range of 0–1000 rpm. An accelerometer mounted on the cutter shaft bearing housing continuously collects vibration acceleration signals. This accelerometer can be a piezoelectric accelerometer. A speed sensor mounted on the frame continuously collects the traveling speed of the straw returning machine. This speed sensor can be a radar speed sensor or a Doppler ultrasonic sensor.

[0028] The torque sensor, speed sensor, acceleration sensor, and velocity sensor all acquire data synchronously at a sampling frequency of 100 Hz, and the data is transmitted to the vehicle controller via the controller area network bus. The sampling frequency can be adjusted within the range of 50–200 Hz as needed.

[0029] In one embodiment of the present invention, the data collection is not continuous throughout the entire operation, but only initiated during the stable operating period of the straw returning machine. The logic for determining the stable operating period is as follows: The controller monitors the travel speed and cutter shaft speed in real time and calculates their real-time rates of change. The rate of change of travel speed is the ratio of the speed difference within adjacent preset time intervals (e.g., 0.5s) to the time interval, with a preset speed fluctuation threshold set at 0.1m / s. 2 The rate of change of the cutter shaft speed is taken as the ratio of the speed difference within an adjacent 0.5s to 0.5s, with a preset speed fluctuation threshold of 5 rpm / s. The controller performs the following judgment at a frequency of 10Hz: If the absolute value of the current travel speed change rate is less than the speed fluctuation threshold, and at the same time the absolute value of the cutter shaft speed change rate is less than the speed fluctuation threshold, then the current moment is determined to be a stable operating period, and sensor data is allowed to enter the buffer and participate in subsequent calculations. If either of the two conditions is not met (e.g., the unit is in an unstable operating condition such as starting, stopping, turning, soil embankment impact, cutter shaft stall, sudden change in cutting resistance, or abnormal fluctuation of cutter shaft speed), then it is determined to be an unstable operating period, the controller suspends data acquisition, clears the data in the current buffer, and resumes acquisition only when both conditions are met simultaneously.

[0030] The above-mentioned dynamic gating mechanism can effectively eliminate abnormal data interference under non-steady-state conditions and significantly reduce the probability of false alarms caused by turning, starting and other operations.

[0031] S3: Calculation of equivalent load characteristic value: In one embodiment of the present invention, the equivalent load characteristic value is calculated as follows: Multiple sets of actual torque and actual speed values ​​are continuously collected within a preset time window. The time window length is set to 2 seconds (it can also be set to 3 seconds or 5 seconds). Within this time window, the controller continuously collects actual torque and actual speed values ​​at a sampling frequency of 100Hz, thus obtaining a total of 200 sets of data within the 2-second window. For each set of data, the instantaneous ratio is calculated, which is the actual torque value divided by the actual speed value at the corresponding moment. Physically, this can be understood as the load torque corresponding to a unit speed of the cutter shaft.

[0032] After obtaining multiple instantaneous ratios, the controller sorts these ratios from smallest to largest. The value in the middle of the sorted sequence is taken as the median. If the data volume is even, the average of the two middle numbers is taken as the median; if it is odd, the exact middle number is taken. The controller uses this median as the equivalent load characteristic value within the time window and uses it as the input to the subsequent neural network.

[0033] The reason for using the median instead of the arithmetic mean is that instantaneous ratios may contain outliers that are abnormally large or small due to accidental impacts from rocks, sensor spike noise, etc. The median is not sensitive to outliers and can more robustly reflect the average resistance level of the tool during normal cutting.

[0034] S4: Extraction of vibration feature values: First, determine the rotational frequency of the toolshaft. This frequency can be calculated from the reading of the speed sensor: Rotational frequency (Hz) = Speed ​​(rpm) / 60. For example, if the current toolshaft speed is 180 rpm, then the rotational frequency is 3 Hz, and its reciprocal, i.e., the period, is approximately 0.333 s.

[0035] In one embodiment of the present invention, the vibration feature values ​​are extracted as follows: Using the reciprocal of the tool shaft rotation frequency as the period, continuously extract at least three complete cycles of sub-signals from the vibration acceleration signal (e.g., three cycles with a total duration of approximately 1 second, or five cycles with a total duration of approximately 1.667 seconds). The starting point for extraction is chosen at the zero-crossing or peak point of the vibration signal to ensure that the waveforms of each cycle are aligned. Perform time-domain synchronous averaging on the extracted three or five cycles of sub-signals: first, align the waveform of each cycle along the time axis, then add the amplitudes at each corresponding time point, and divide by the number of cycles extracted to obtain an averaged single-cycle vibration waveform. This averaging process can effectively eliminate random noise that is out of sync with the tool shaft rotation frequency and non-periodic interference from other components (such as the engine and drivetrain), thereby enhancing the periodic impact component related to tool wear.

[0036] Then, a Fast Fourier Transform (FFT) is performed on the averaged single-cycle vibration waveform, with 1024 points used for the transform. The formula for the Fast Fourier Transform is as follows: ; in, Represents the amplitude in the frequency domain; Represents the time-domain sampling sequence, and represents the first... The amplitude of each sampling point; This represents the index of the time-domain sampling point, with a value range of 1. ; Represents the natural constant; Represents the imaginary unit; This represents the number of transformation points; in this embodiment, N=1024. This represents the phase factor in the discrete Fourier transform.

[0037] From the transformed spectrum, extract the amplitude corresponding to the fundamental frequency position of the tool shaft rotation frequency (e.g., the spectral line amplitude at 3Hz) and the amplitude corresponding to the second harmonic (i.e., 2 times the fundamental frequency) position. Combine the fundamental frequency amplitude and the second harmonic amplitude into a vibration characteristic value according to the following formula: ; in, This indicates the amplitude corresponding to the fundamental frequency; Indicates the second harmonic amplitude; This represents the synthesized vibrational characteristic value.

[0038] This vibration characteristic value directly reflects the magnitude of the periodic impact force generated by the wear, chipping, or deformation of the cutting edge during each revolution of the cutter shaft. The fundamental frequency amplitude mainly reflects the periodic impact energy generated by the interaction between the cutter and the soil during each revolution; the second harmonic amplitude is more sensitive to the imbalance of rotating components, and its increase may originate from various factors such as uneven wear of the cutter, local defects, eccentricity of the cutter shaft assembly, or changes in bearing clearance. By first capturing multiple complete cycles and then performing time-domain synchronous averaging, the random noise that is out of sync with the cutter shaft rotation frequency is effectively reduced, and the periodic impact component directly related to cutter wear is enhanced, thereby improving the sensitivity of the vibration characteristics to the cutter wear state.

[0039] S5: Neural network model predicts wear probability: The equivalent load characteristic value, vibration characteristic value, surface soil electrical conductivity value, surface soil moisture content value, travel speed value, load-vibration coupling coefficient, and salt-velocity erosion index are input into a pre-trained neural network model, and the wear degree probability value is output.

[0040] Among them, the load-vibration coupling coefficient is constructed based on the equivalent load characteristic value and the vibration characteristic value, and the salt-velocity erosion index is constructed based on the surface soil electrical conductivity value and the travel speed value.

[0041] In one embodiment of the present invention, such as Figure 2 As shown, the neural network model adopts a multi-branch structure, specifically including a feature input layer, a first branch, a second branch, a third branch, and a fusion output layer.

[0042] Feature input layer: The feature input layer first performs quantile normalization on each of the five input parameters. The formula for quantile normalization is as follows: ; in, This represents the normalized eigenvalues, ranging from 0 to 1; Represents the original data value; This represents the selected lower quantile in the original data distribution, such as the 10th quantile or the 1st quantile. This represents the selected upper quantile in the original data distribution, such as the 90th quantile. The specific process is as follows: For each input parameter (equivalent load characteristic value, vibration characteristic value, surface soil electrical conductivity value, surface soil moisture content value, and travel speed value), at least 1000 samples are collected in advance from historical operation data. 100 quantile points (i.e., 1% quantile, 2% quantile, ..., 99th quantile) are calculated for these samples to form a quantile mapping table. When a new measured value arrives, the feature input layer determines between which two adjacent quantiles the measured value falls into, and then maps it to a quantile value between 0 and 1 through linear interpolation.

[0043] Based on the five normalized values, the feature input layer further generates two cross features. For example... Figure 2 middle" "Torque × Vibration" and " The label "Conductivity × Velocity" indicates that... "" indicates a multiplication operation of corresponding features, that is, a cross-combination of two input features to generate a new feature. Based on this, the present invention constructs the following two cross-features: The first cross feature is obtained by multiplying the normalized equivalent load feature value with the normalized vibration feature value to obtain the load-vibration coupling coefficient. The value of this coefficient is also between 0 and 1. The larger the value, the higher the load and vibration are at the same level. This usually occurs when the tool wear is severe, which leads to increased cutting resistance and aggravated rotational imbalance.

[0044] The second cross-feature involves multiplying the normalized surface soil conductivity value by the normalized travel speed value to obtain the salt-velocity erosion index. This index characterizes the accelerated erosion effect of high-speed travel on the cutting tool in high-salt soil environments. Its physical significance lies in the fact that the faster the travel speed, the greater the number of contacts and friction distance between the cutting tool and the saline soil per unit time, thus amplifying the corrosive effect of salt.

[0045] First branch: The first branch is a three-layer fully connected neural network with the following structure: The input layer has 5 nodes, receiving normalized equivalent load feature values, normalized vibration feature values, normalized surface soil electrical conductivity values, normalized surface soil moisture content values, and normalized travel velocity values, respectively. The first hidden layer has 16 nodes, and the activation function is the ReLU function (Rectified Linear Unit), as shown in the following formula: ; in, This indicates that the output is activated (taking the maximum value between 0 and z). This represents the weighted input value of the neuron.

[0046] The second hidden layer has 8 nodes, and the activation function is also ReLU. The output layer has 1 node, and the activation function is Sigmoid. The first branch outputs the first wear probability value. The network is trained using the backpropagation algorithm, with a binary cross-entropy loss function.

[0047] Second branch: The second branch is a logistic regression classifier, which takes two input features: the load-vibration coupling coefficient and the salinity-velocity erosion index. The expression for the logistic regression model is as follows: ; in, This represents the second wear probability value output by the second branch; and These represent two input features (i.e., load-vibration coupling coefficient and salt-velocity erosion index). and These represent the weighting coefficients; Indicates the bias term; Represents the natural constant.

[0048] The above parameters are obtained through training, and the second branch outputs the second wear probability value. .

[0049] Third branch: The third branch is also a logistic regression classifier, whose input features are only the normalized surface soil electrical conductivity and normalized surface soil moisture content, and its output is the third wear probability value. This branch is used to detect the tendency of tools to corrode and wear solely due to soil environment (salt and moisture).

[0050] The fusion output layer is an adaptive discreteness fusion layer, and its fusion rules are as follows: First, calculate the first wear probability value. Second wear probability value and the third wear probability value range The range is defined as the maximum value minus the minimum value: ; in, Indicates taking The maximum value in; Indicates taking The minimum value in.

[0051] Preset a discreteness threshold The dispersion threshold can be set to 0.1–0.2, preferably 0.15. The dispersion threshold used to determine whether the outputs of the three branches are consistent is set when the range of the three probability values ​​is less than 1. If all branches agree, it is considered that there is a significant disagreement; otherwise, it is considered that there is a significant disagreement. (Comparison of fusion layers) With threshold The size relationship is determined by the following merging rules: ; in, This represents the probability value of the final wear level; Indicates taking the median; This represents the dispersion threshold (a dimensionless constant, 0.1 to 0.2).

[0052] when At this point, if the judgments of the three branches are considered to be relatively consistent and without significant disagreement, then... , , The arithmetic mean of these values ​​is used as the final probability value for the degree of wear. When If the outputs of the three branches show significant divergence, it may be due to abnormal input or noise interference in one of the branches. In this case, take... , , The median was used as the final probability value for wear level to suppress the influence of outliers.

[0053] Training methods for neural network models: (1) Training Sample Collection: Selected work plots with different degrees of salinization and started work using brand new cutting tools. The machine was stopped at fixed intervals when the cumulative cutting work was fixed or when obvious signs of wear appeared, and the wear status label of the cutting tool was obtained. The wear status label was determined based on one or more of the following indicators: cutting edge wear width (VB value), cutting tool mass loss rate, straw crushing length qualification rate or burial coverage rate after operation. Specifically, the cutting edge wear width VB value was the main basis. When VB ≥ 1.5 mm, it was marked as positive (needs to be replaced), and when VB < 1.5 mm, it was marked as negative (can continue to be used). The torque, rotational speed, vibration, velocity and soil parameters during this period were recorded simultaneously as a training sample.

[0054] (2) Model Training: 80% of the training set is used for parameter updates, and 20% for validation. The multi-branch network is trained end-to-end using the binary cross-entropy loss function and the Adam optimizer until the validation set accuracy no longer improves. The formula for the binary cross-entropy loss function is as follows: ; in, This represents the loss function value, which measures the difference between the model's predictions and the true labels. This represents the number of training samples, and this represents the total number of samples participating in the training. Indicates to The loss values ​​of each sample are averaged. Represents the sample index, indicating the first... There are 1 training samples, with values ​​ranging from 1 to 1. ; Indicates the first The true label of each sample; Represents the natural logarithm; This represents the probability value of the wear level output by the model; The loss term represents the loss term for positive class samples; This represents the loss term for negative class samples.

[0055] This loss function is used to quantify the difference between the wear probability value predicted by the model and the actual wear label. By minimizing this loss value, the network parameters are updated so that the wear probability output by the model gradually approaches the actual wear state.

[0056] S6: Determination of tool wear condition: The controller outputs the wear probability value from the neural network model. Compare with a preset threshold. When When the wear exceeds a preset threshold, the controller determines that the current tool wear has reached the point where it needs to be replaced. The controller will issue an alarm prompt through the vehicle display screen or issue an audible alarm through the speaker.

[0057] In one embodiment of the present invention, the preset threshold is a dynamic threshold, which is determined as follows: During each operation, the controller uses the actual torque value collected by the torque sensor. and the actual rotational speed value collected by the rotational speed sensor Real-time calculation of instantaneous cutting power Instantaneous cutting power is used to quantify the rate at which the cutting energy is consumed by the tool at the current moment, and is the basis for calculating cumulative cutting power. The calculation formula is as follows: ; in, Indicates instantaneous cutting power, in kW; This represents the actual torque value, in N·m. This represents the actual rotational speed, in rpm. This represents a preset coefficient used to convert the units of torque and speed to the units of power. Its physical basis is 9550≈60×1000 / (2π), which means converting the speed from rpm to rad / s and the power from W to kW.

[0058] The controller uses a fixed sampling time interval. Accumulate points. Select 0.1 s. From the initial moment when the tool is installed on the machine and begins use, each instantaneous power Multiply The infinitesimal cutting work done during this time period is obtained, and then added to the previous sum to obtain the total cumulative cutting work from the initial tool usage time to the current time. Cumulative cutting work reflects the total cutting workload completed by the tool from the start of its use to the current moment, and is an important indicator of the cumulative wear of the tool. The formula for calculating cumulative cutting work is as follows: ; in, This represents the total cumulative cutting work, expressed in kJ. This indicates the sampling time interval, measured in seconds (s).

[0059] Total cumulative cutting work The values ​​are stored in the controller's non-volatile memory, and are retained after each job is completed, and are continued to accumulate in the next job.

[0060] Then, the controller will calculate the total accumulated cutting power. The input is fed into a pre-calibrated threshold decay function to obtain the preset threshold at the current time. For example... Figure 3 As shown, the threshold decay function is an exponential decay function, used to simulate the physical law of tool life gradually decreasing due to wear accumulation during cumulative cutting, so that the alarm threshold adaptively decreases as the actual workload of the tool increases. The form of the threshold decay function is as follows: ; in, This represents the dynamic threshold at the current moment, which is the threshold for determining the probability that the tool needs to be replaced. The initial threshold is set to 0.75, which represents the baseline value for determining the probability that a new tool needs to be replaced before wear occurs. Represents the natural constant; Expressed as an exponential decay factor, it represents the cumulative cutting work. Under the influence of the current threshold relative to the initial threshold, the rate of decay of the current threshold gradually decreases as the cumulative cutting work E increases, thus causing the threshold to exhibit an exponential decay trend.

[0061] The controller will be at the current moment As a preset threshold, it is used to compare with the final wear probability value output by the neural network. A comparison was made. The threshold decay function was calibrated by performing nonlinear least squares fitting on at least five sets of full-life-cycle data of the same type of tool, with an initial threshold set during the fitting process. This is a reasonable alarm threshold for new tooling conditions, in order to avoid too many false alarms due to an excessively low threshold at the end of the tool's lifespan.

[0062] In one embodiment of the present invention, to improve the reliability of the determination, instead of relying on the comparison between the wear probability value of a single sample and the threshold, a multi-time voting mechanism within a sliding time window is adopted. For example... Figure 4 As shown, this voting mechanism is used to filter out over-threshold events caused by occasional noise or transient disturbances. An alarm is only triggered when the wear condition is repeatedly confirmed over a continuous period. Specifically: The controller maintains a first-in, first-out (FIFO) queue in memory. The maximum storage capacity of the queue is L (L can be chosen as 5, 10, or 15), representing the wear probability values ​​corresponding to L consecutive sampling times within the window. Each time the neural network model outputs a new wear probability value, the controller pushes that value to the tail of the queue; if the number of elements in the queue exceeds L, the oldest value at the head of the queue is popped. The queue always retains the probability values ​​of the most recent L samples after reaching full capacity.

[0063] Simultaneously, the controller sets a trigger threshold K (K can be 3, 6, or 8), requiring K ≤ L. At each sampling time, the controller counts how many probability values ​​in the current queue are greater than the current preset threshold. When the statistical result is greater than or equal to K, that is, when at least K probability values ​​in the queue are greater than the threshold, the controller finally determines that the tool wear condition is in a state that needs to be replaced and triggers a replacement alarm signal.

[0064] For example, with L=5 and K=3, the five probability values ​​in the queue are 0.72, 0.68, 0.73, 0.71, and 0.69, respectively. The threshold is 0.70. The number of values ​​greater than the threshold is 3, reaching K, therefore a replacement is determined. If only one or two probability values ​​are greater than the threshold, it is considered a transient disturbance or occasional noise, and a replacement determination is not triggered. The sliding time window continuously moves forward with the arrival of new sampled values, continuously performing the above judgment.

[0065] The majority voting mechanism within the sliding time window effectively filters out isolated and occasional over-threshold events. An alarm is only issued when the wear status is confirmed multiple times over a period of time, significantly reducing the false alarm rate.

[0066] It is understood that the above embodiments are merely exemplary implementations used to illustrate the principles of the present invention, and the present invention is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and essence of the present invention, and these modifications and improvements are also considered to be within the scope of protection of the present invention.

Claims

1. A method for identifying the wear condition of blades in a straw returning machine for saline-alkali land, characterized in that, Includes the following steps: S1, obtain the surface soil electrical conductivity and surface soil moisture content of the current working site; S2, during the operation of the straw returning machine, collect the actual torque value of the power output shaft, the actual rotation speed value of the cutter shaft, the vibration acceleration signal at the cutter shaft bearing seat, and the travel speed value of the straw returning machine; S3, calculate the equivalent load characteristic value based on the actual torque value and the actual speed value; S4, perform frequency domain transformation on the vibration acceleration signal, and extract the amplitude corresponding to the fundamental frequency and the second harmonic amplitude of the tool shaft rotation frequency as vibration characteristic values; S5, the equivalent load characteristic value, the vibration characteristic value, the surface soil electrical conductivity value, the surface soil moisture content value, the travel speed value, the load-vibration coupling coefficient, and the salt-velocity erosion index are input into a pre-trained neural network model, and the wear degree probability value is output; wherein, the load-vibration coupling coefficient is constructed based on the equivalent load characteristic value and the vibration characteristic value, and the salt-velocity erosion index is constructed based on the surface soil electrical conductivity value and the travel speed value; The neural network model includes: The feature input layer is used to normalize the input equivalent load feature value, vibration feature value, surface soil electrical conductivity value, surface soil moisture content value and travel speed value respectively, and generate the load-vibration coupling coefficient and the salt-velocity erosion index based on the normalized values. The first branch is used to input the normalized equivalent load characteristic value, the vibration characteristic value, the surface soil electrical conductivity value, the surface soil moisture content value, and the travel speed value, and output the first wear probability value. The second branch is used to input the load-vibration coupling coefficient and the salt-velocity erosion index, and output the second wear probability value. The third branch is used to input the normalized surface soil electrical conductivity value and the surface soil moisture content value, and output the third wear probability value. A fusion output layer is used to fuse and output a final wear probability value based on the first wear probability value, the second wear probability value, and the third wear probability value. S6. When the wear probability value is greater than a preset threshold, the tool wear condition is determined to be in a state that needs to be replaced.

2. The method for identifying the wear state of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, In step S1, multiple sampling points are selected in a grid or diagonal pattern within the current work area. The surface soil electrical conductivity and surface soil moisture content of each sampling point are measured, and the location coordinates of each sampling point are recorded. The location coordinates of each sampling point and its corresponding electrical conductivity and moisture content are stored as a plot parameter lookup table. During the operation of the straw returning machine, the current working position is obtained through the positioning device. Based on the spatial distance between the current working position and the sampling point positions in the plot parameter lookup table, the conductivity value and moisture content value corresponding to the nearest sampling point are selected as the conductivity value and moisture content value of the surface soil.

3. The method for identifying the wear condition of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, In step S2, the acquisition of the actual torque value, the actual rotational speed value, the vibration acceleration signal, and the travel speed value are all carried out during the stable operation period of the straw returning machine; the stable operation period is determined by the real-time change rate of the travel speed value being less than a preset speed fluctuation threshold and the real-time change rate of the actual rotational speed value being less than a preset rotational speed fluctuation threshold.

4. The method for identifying the wear condition of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, In step S3, multiple sets of actual torque values ​​and actual speed values ​​are continuously collected within a preset time window. The instantaneous ratio of each set of actual torque values ​​to the corresponding actual speed values ​​is calculated, and the median of multiple instantaneous ratios is taken as the equivalent load characteristic value.

5. The method for identifying the wear state of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, In step S4, at least three complete cycles of sub-signals are continuously extracted from the vibration acceleration signal with the reciprocal of the rotation frequency of the cutter shaft as the period. The extracted sub-signals are then synchronously averaged in the time domain to obtain the averaged single-cycle vibration waveform. The averaged single-cycle vibration waveform is then subjected to the frequency domain transformation to extract the amplitude corresponding to the fundamental frequency and the amplitude of the second harmonic, which are used as the vibration characteristic values.

6. The method for identifying the wear condition of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, The fusion output layer calculates the range of the first wear probability value, the second wear probability value, and the third wear probability value; when the range is less than a preset dispersion threshold, the arithmetic mean of the first wear probability value, the second wear probability value, and the third wear probability value is used as the wear degree probability value. When the range is greater than or equal to the preset dispersion threshold, the median of the first wear probability value, the second wear probability value, and the third wear probability value is taken as the wear degree probability value.

7. The method for identifying the wear condition of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, During each operation, based on the actual torque value and the actual rotational speed value, the instantaneous cutting power is calculated by dividing the instantaneous power by the product of the torque and the rotational speed by a preset coefficient. The instantaneous cutting power is integrated over time to obtain the total cumulative cutting power from the initial use of the tool to the current time. The total cumulative cutting power is input into a pre-calibrated threshold decay function to obtain the preset threshold at the current time. The threshold decay function is an exponential decay function.

8. The method for identifying the wear condition of blades in a straw returning machine for saline-alkali land according to claim 1, characterized in that, In step S6, a sliding time window is set, which contains the wear probability value corresponding to L consecutive sampling times. When the wear probability value of at least K sampling times in the sliding time window is greater than the preset threshold, the tool wear state is determined to be a state that needs to be replaced. Here, L is the total number of sampling times in the window, K is the preset trigger threshold, and K≤L.

Images