A peanut harvester working condition detection and control method and system

Abstract

The present application belongs to the field of agricultural machinery automation technology, and particularly relates to a peanut harvester working condition detection and control method and system, which comprises the following steps: obtaining low-frequency impedance real part mean value and high-frequency phase angle mean value through multi-frequency impedance measurement as conductive capacity measurement and polarization response index; calculating real part mean value adjacent difference mean value and phase angle lag time autocorrelation coefficient in a sliding window, and determining a humidity softening event according to comparison of the two with threshold values; when it is determined that a humidity softening event occurs, searching for similar working conditions from a historical working condition case library and calculating a target depth compensation value by reciprocal weighting of fruit breaking rate; combining hydraulic actuator lag characteristics to output a depth correction instruction in advance; and updating an adaptive compensation coefficient based on online fruit breaking rate feedback to adjust the instruction. The present application uses double feature logic to distinguish between humidity changes and interference, accurately captures humidity mutations, realizes predictive depth control and online learning, and reduces fruit breaking rate.

Description

Technical Field

[0001] This invention relates to the field of agricultural machinery automation technology. More specifically, this invention relates to a method and system for detecting and controlling the operating conditions of a peanut harvester. Background Technology

[0002] In peanut harvesting operations, uneven spatial distribution of soil moisture content can lead to an increased rate of broken peanuts. Therefore, real-time monitoring of soil moisture and adjustment of digging depth based on the monitoring results are effective means to reduce losses.

[0003] Existing technologies typically employ contact soil impedance measurement methods to obtain moisture information. This involves measuring the amplitude or phase angle of the soil complex impedance using a fixed frequency excitation method, establishing a correlation between the changing trend of a single parameter and moisture, and then driving the adjustment of excavation depth.

[0004] However, when harvesters are operating in the field, non-humidity factors such as mechanical vibration, poor instantaneous contact between electrodes and soil, temperature drift, and electrode surface passivation can couple into the impedance signal in different forms. This results in a large number of interferences unrelated to humidity mixing into the measurement results of a single frequency or single parameter. For example, the impact of stones can cause short-term spikes in the impedance amplitude, while electrode passivation can cause a slow unidirectional drift in the phase angle. If these interferences are not identified, they can easily be misjudged as changes in soil moisture, thus leading to improper adjustment of digging depth.

[0005] How to reliably identify sudden changes in soil moisture and overcome these challenges has become a pressing technical problem in this field. Summary of the Invention

[0006] To address the technical problems of hydraulic lag, deviation in digging depth adjustment, and high breakage rate, this invention provides solutions in the following aspects.

[0007] In a first aspect, the present invention provides a method for detecting and controlling the working condition of a peanut harvester, which adopts the following technical solution: a method for detecting and controlling the working condition of a peanut harvester, comprising: performing multi-frequency impedance measurement on the soil, determining the conductivity measure based on the real part of the impedance in the low-frequency band, and determining the polarization response index based on the phase angle in the high-frequency band, comprising: arranging multiple annular electrodes at equal intervals on a crossbeam perpendicular to the direction of travel below the front edge of the harvester's digging shovel. Impedance excitation and acquisition are performed sequentially at multiple frequencies of a preset logarithmic distribution sequence; The average real part of the impedance at multiple frequency points in the low-frequency band is taken as the conductivity measure, and the average phase angle at multiple frequency points in the high-frequency band is taken as the polarization response index. Within a sliding window, the mean of the absolute differences between adjacent sampling points of the conductivity measure and the autocorrelation coefficient of the polarization response index under a preset lag time are calculated, including: a preset sliding window and a preset lag time; for the conductivity measure sequence within the sliding window, the arithmetic mean of the absolute differences between adjacent sampling points is calculated to obtain the mean of the absolute differences between adjacent sampling points of the conductivity measure; from the original sequence of the polarization response index, two sub-sequences are obtained by offsetting according to the preset lag time, and the Pearson correlation coefficient of the two sub-sequences is calculated to obtain the autocorrelation coefficient of the polarization response index under the preset lag time; When the mean value is greater than the first threshold and the autocorrelation coefficient is less than the second threshold, a humidity softening event is determined to have occurred. In response to the humidity softening event, similar operating conditions are retrieved from the historical operating condition case library, and the target depth compensation value is obtained by weighting the inverse of the fruit breakage rate. Based on the difference between the target depth compensation value and the current excavation depth, and combined with the hysteresis characteristics of the hydraulic actuator, a depth correction command is obtained and output in advance; Furthermore, the adaptive compensation coefficient is updated based on the online fruit breakage rate feedback, and the amplitude of the subsequent depth correction command is adjusted using the updated adaptive compensation coefficient.

[0008] Preferably, the mean value is greater than a first threshold and the autocorrelation coefficient is less than a second threshold, including: the first threshold is an upper bound of the conductivity deflection gradient set based on the normalization of the impedance amplitude of the dry region; the second threshold is a lower bound of the phase angle lag autocorrelation coefficient.

[0009] Preferably, the step of retrieving similar working conditions from the historical working condition case library and obtaining the target depth compensation value by weighting the reciprocal of the failure rate includes: each record in the historical working condition case library includes a conductivity measure, a polarization response index, a digging depth correction value, and a failure rate; calculating the Mahalanobis distance between the query vector composed of the current conductivity measure and polarization response index and the feature vectors of each record in the historical working condition case library; determining a preset number of records with the smallest Mahalanobis distance as the similar working conditions; and performing a weighted average of the digging depth correction values ​​corresponding to the similar working conditions, using the reciprocal of the failure rate plus a preset non-zero small constant as the weight, to obtain the target depth compensation value.

[0010] Preferably, the step of obtaining a depth correction command and outputting it in advance based on the difference between the target depth compensation value and the current excavation depth, combined with the hysteresis characteristics of the hydraulic actuator, includes: obtaining the inertial time constant and pure hysteresis time in the first-order inertial plus pure hysteresis model of the hydraulic actuator; calculating the advance time as the pure hysteresis time plus twice the inertial time constant; generating a step-form depth correction command based on the difference between the target depth compensation value and the current excavation depth, and issuing it in advance by the advance time to compensate for the hysteresis of the hydraulic actuator.

[0011] Preferably, the step of updating the adaptive compensation coefficient based on online fruit breakage rate feedback and adjusting the amplitude of the subsequent depth correction command using the updated adaptive compensation coefficient includes: obtaining the estimated fruit breakage rate of the current travel segment and the estimated fruit breakage rate of the previous travel segment; calculating the compensation coefficient update amount using the sliding gradient descent method based on the difference between the current value and the previous value of the adaptive compensation coefficient, combined with the change in the estimated fruit breakage rate, and updating the adaptive compensation coefficient.

[0012] Preferably, the depth correction instruction includes: setting a swing reversal coefficient based on the comparison result between the target depth compensation value and the current excavation depth; setting the swing reversal coefficient to be greater than 1 when the target depth is less than the current depth and the excavation depth needs to be reduced; setting the swing reversal coefficient to be less than 1 when the target depth is greater than the current depth and the excavation depth needs to be increased, and multiplying the step-form depth correction instruction by the swing reversal coefficient.

[0013] Secondly, the present invention provides a peanut harvester condition detection and control system, which adopts the following technical solution: a peanut harvester condition detection and control system includes a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned peanut harvester condition detection and control method is implemented.

[0014] The present invention has the following beneficial effects: 1. By simultaneously calculating the cumulative deflection gradient of adjacent samples of conductivity metric and the lag autocorrelation coefficient of the phase angle sequence within a sliding time window, and determining humidity softening events based on the combination of these two conditions, the discrimination criterion can be anchored to the composite feature of a rapid jump in the real part of impedance and instantaneous instability of the phase sequence occurring simultaneously during a sudden change in water content. The conductivity gradient exceeding the limit is responsible for capturing the sudden drop in resistance caused by humidity, while the phase autocorrelation lower limit eliminates isolated spike artifacts caused by mechanical vibration or instantaneous electrode open circuits. The synergy of these two features makes the discrimination logic immune to both transient shocks and gradual drift, reducing the probability of false triggering caused by rock impacts and bumps, and helping to ensure that excavation depth adjustment commands are issued only at the moment of a true humidity change.

[0015] 2. By introducing a historical case library of working conditions and using a weighted index of the inverse of the fruit breakage rate to identify similar working conditions, the current depth compensation decision inherits historical low-loss control experience, forming a feedforward reference optimal value, replacing the rigid adaptation of fixed mapping relationships to varying soil conditions; at the same time, based on the first-order inertia plus pure lag model of hydraulic cylinders, the depth correction command is output in advance, so that the actual action of the shovel blade arrives synchronously with the wet zone boundary, systematically compensating for the adjustment delay and overshoot caused by hydraulic lag, making the adjustment process smoother; further supplemented by an adaptive gradient descent mechanism with online fruit breakage rate feedback, the command gain coefficient is continuously adjusted to form a feedforward-feedback closed optimization, enabling the system to automatically converge the control intensity according to soil conditions and equipment wear, achieving a sustained low level of fruit breakage rate under long-term field operations. Attached Figure Description

[0016] Figure 1 The schematic diagram illustrates a method flowchart of a peanut harvester operating condition detection and control method and system according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] Step S1: Perform multi-frequency impedance measurement to obtain conductivity measurement and polarization response index.

[0020] During the operation of peanut harvesters in the field, the ring electrode array installed on the crossbeam in front of the front edge of the digging shovel is in continuous contact with the soil, forming a soil-electrode impedance measurement circuit.

[0021] The horizontal distance between the crossbeam and the digging shovel is 0.5 meters in the direction of the harvester's travel to ensure sufficient time for action after detecting a moisture softening event. The annular electrodes are made of 316L stainless steel, with an inner diameter of 10 mm and an outer diameter of 16 mm. The center-to-center distance between adjacent electrodes is 20 cm, and there are a total of 4 electrodes. The electrodes are isolated from the crossbeam by a polytetrafluoroethylene insulating sleeve. The working surface of the electrodes maintains a contact pressure of 2 Newtons with the soil, which is achieved by a spring clamping mechanism. The sensor selection and installation parameters can be adjusted by the implementer according to the specific implementation scenario.

[0022] Because soil moisture content is extremely unevenly distributed in space, and non-humidity factors such as mechanical vibration during harvesting, poor instantaneous contact between electrodes and soil, and temperature drift occur, directly measuring soil impedance using a single frequency or single impedance parameter results in a large number of interference components unrelated to humidity changes in the measurement results.

[0023] For example, when a stone hits the electrode and causes a momentary circuit break, the impedance amplitude will produce a short-term spike jump; while when the electrode surface is slowly passivated due to continuous operation, the phase angle will show a unidirectional slow drift.

[0024] If these non-humidity interferences are introduced into subsequent judgments without differentiation, it is very easy to cause misjudgment of the soil moisture status.

[0025] Therefore, by using multi-frequency impedance measurement and extracting two characteristic quantities with different focuses from the frequency domain differences, we can reflect the soil volume conductivity and electrode interface polarization characteristics respectively, so as to preserve the differential response information of the soil-electrode system under low-frequency and high-frequency excitation.

[0026] Below the front edge of the harvester's digging shovel, along a crossbeam perpendicular to the direction of travel, four stainless steel ring electrodes are arranged at equal intervals to form a near-body soil impedance sensor array.

[0027] Each pair of electrodes forms a measurement pair, which are then sequentially connected to the embedded impedance analysis module via a switching circuit.

[0028] This module contains a digital frequency synthesizer and a quadrature lock-in amplifier, which are used to generate excitation signals and synchronously acquire response signals.

[0029] The excitation frequency adopts a logarithmic distribution sequence, with a total of eight frequency points selected. The lowest frequency is 1 kilohertz, the highest frequency is 100 kilohertz, and the intermediate frequency points are determined by logarithmic intervals. The frequency selection can be adjusted by the implementer according to the specific implementation scenario.

[0030] For each frequency point, a sinusoidal AC excitation current is sequentially injected into the measuring electrode pair, and the amplitude of the response voltage and its phase difference relative to the excitation current are collected. The complex impedance at that frequency is obtained through quadrature lock-in amplification. ,in, The real part of the impedance. This is the imaginary part of the impedance. The imaginary unit, Indicates the first The excitation frequency is measured in Hertz.

[0031] To suppress the nonlinear distortion introduced by the electrode-soil interface polarization effect in the high-frequency band, the measured complex impedance at each frequency point is calibrated by least squares fitting to obtain the calibrated impedance real part sequence and impedance imaginary part sequence. However, the implementer can also directly use the original real part and imaginary part according to the field conditions.

[0032] After obtaining the calibrated real and imaginary parts, the corresponding phase angle is calculated for each frequency point. The unit is radians.

[0033] To suppress nonlinear errors caused by electrode polarization, sampling begins after 5 cycles of excitation at each frequency point, with a sampling window of 10 cycles. The sampling results are then subjected to median filtering with a window length of 3 to remove transient spikes caused by stone impacts. The number of excitation cycles, the number of sampling cycles, and the size of the median filtering window are all empirical values ​​that can be adjusted by the implementer according to the specific implementation scenario.

[0034] Because low-frequency excitation current can penetrate deeper into the soil, the real part of the impedance of the measurement circuit is mainly dominated by the ionic conductivity of the soil, with little contribution from interface polarization. However, under high-frequency excitation, the charge accumulation and relaxation process at the electrode-soil interface is significantly enhanced, and the phase angle is more sensitive to changes in polarization state.

[0035] Therefore, by utilizing this difference in frequency domain response, the changes in bulk conductivity and interface polarization caused by humidity variations can be mapped onto the low-frequency real part and the high-frequency phase angle, respectively.

[0036] Therefore, three frequencies within the low-frequency band are selected from the eight frequencies. to The corresponding frequency range is approximately 1 kHz to 5 kHz. The arithmetic mean of the real part of the calibrated impedance is taken as the measure of conductivity. .

[0037] At the same time, three frequency points within the high-frequency band were selected, namely to The corresponding frequency range is approximately 20 kHz to 100 kHz. The arithmetic mean of the phase angles is taken as the polarization response index. , dimensionless.

[0038] The calculation method involves summing the real parts of the impedances at the three low-frequency points and dividing by three to obtain the conductivity measure. The unit is ohms; the polarization response index is obtained by summing the phase angles at the three frequency points in the high-frequency band and dividing by three. , dimensionless.

[0039] Within each impedance sampling period of 0.1 seconds, the above process completes one full frequency sweep and mean calculation, outputting a two-dimensional feature pair consisting of a conductivity metric and a polarization response index. .

[0040] This non-orthogonal feature extraction method eliminates the need to forcibly decouple complex impedance into pure resistive and pure capacitive components. Instead, it directly utilizes observations with different physical emphases in the low-frequency and high-frequency domains to form conductivity metrics and polarization response indices, respectively.

[0041] Conductivity measurement mainly characterizes the ability of ions to migrate in the volume pathway of soil pore water. When soil moisture content increases significantly, the ion migration channels widen, and the value of this measurement decreases. On the other hand, polarization response index mainly characterizes the polarization relaxation state of the electrode-soil interface. When a sudden change in soil moisture causes a rapid reconstruction of the ion distribution at the interface, this index shows short-term decorrelation fluctuations.

[0042] Step S2: Calculate the cumulative deflection gradient of conductivity and the autocorrelation coefficient of polarization response lag within the sliding window.

[0043] Electrical conductivity measurement during harvester field operations The sustained decrease in the sequence was mainly caused by abrupt changes in soil moisture content, while transient disturbances such as rock impacts and poor contact only produced isolated spikes. Therefore, within the sliding window... Calculating the average of the differences between adjacent points can accumulate the true mutation intensity through the arithmetic mean effect and suppress isolated spikes.

[0044] Polarization response index The short-term self-similarity of the sequence decreases significantly due to rapid interface reconstruction during abrupt humidity changes, while slow-change processes such as temperature drift maintain a high autocorrelation. Therefore, the degree of instability of the polarization state can be quantitatively described by calculating the lag autocorrelation coefficient. Thus, this step extracts the cumulative deflection gradient of conductivity within the sliding window. and polarization response lag autocorrelation coefficient Two statistics.

[0045] With sampling period Continuous reception of conductivity measurement and polarization response index and using a time width The sliding observation window has a first-in-first-out (FIFO) queue structure, and the number of sampling points within the window is fixed. Each new set of feature pairs is stored; The earliest set of data is automatically removed, and the window always retains the twenty consecutive sets of sampled values ​​from the most recent two seconds. This sequence is denoted as... and The time span can be adjusted by the implementer according to the specific implementation scenario.

[0046] Calculate the mean of the absolute differences between adjacent sampling points of conductivity measurement within the window, and then calculate the absolute difference for each pair of adjacent points sequentially. ,in , together The cumulative deflection gradient of conductivity is obtained by taking the arithmetic mean of the differences. The calculation formula is:

[0047] In the formula, For the first one inside the window The conductivity of each sampling point is measured in ohms. , Units and The mean is in ohms; this mean combines the average of all local jumps within the window, significantly attenuating isolated spikes due to the arithmetic mean, while retaining higher values ​​for large drops at multiple consecutive points, thus highlighting the continuous change characteristics caused by sudden changes in humidity.

[0048] Calculate the autocorrelation coefficient of the polarization response exponent under a preset time delay, where the number of sampling points corresponding to the preset time delay is [value missing]. That is, time offset The number of sampling points can be adjusted by the implementer according to the specific implementation scenario; the first few points are extracted from the original sequence. The points constitute a subsequence , and afterwards The points constitute a subsequence ,calculate and The Pearson correlation coefficient is used to obtain the polarization response lag autocorrelation coefficient. The calculation formula is as follows:

[0049] In the formula, For the first one inside the window The polarization response exponent at each sampling point is dimensionless. , and Subsequences and The mean, The coefficient is dimensionless, and its range of values ​​is [range missing]. to The closer the coefficient is to zero, the better the polarization response. The weaker the self-similarity over the time scale, the higher the degree of instability, which coincides with the rapid interface reconstruction process caused by sudden changes in humidity, with coefficients close to... This indicates a high degree of autocorrelation between the preceding and following sequences, corresponding to a gradual drift process.

[0050] When the accumulated data points of the sliding window reach At that time, the above and This outputs the dynamic feature indicators corresponding to the current window. During the period after power-on startup when the buffer is not full, this step does not perform calculations and will continue to update after the window is filled.

[0051] Step S3: Determine the humidity softening event based on the gradient and autocorrelation coefficient.

[0052] Accumulate twenty sampling points within the sliding observation window and complete the conductivity accumulation deflection gradient. Autocorrelation coefficient with polarization response lag After calculation, if only As a criterion, single-point impedance spikes caused by transient interference such as stone impacts on electrodes may still exceed normal fluctuation levels after arithmetic averaging, even though they only occupy one or two sampling points within the window. The value caused a false triggering.

[0053] And alone As a criterion, under slow-changing processes such as temperature drift or electrode passivation... It consistently remained at a high value close to 1, making it impossible to correlate with actual humidity fluctuations. The descent is distinguished, therefore this step uses... and The dual-feature logic combination is used for judgment to simultaneously suppress transient mechanical spikes and slowly varying drift components.

[0054] because The original calculated values ​​have ohmic dimensions. Differences in impedance baselines between arid soils in different plots mean that the same gradient threshold lacks universality. Therefore, [the following is omitted as the original text is incomplete and cannot be translated]. Divide by the preset reference impedance value of the drying zone Normalize, The empirical value of 15.0 ohms is adopted. This value can be obtained by automatically collecting the average real part of the impedance in a known dry section after the system is turned on, or it can be manually calibrated by the operator before operation and normalized gradient. Dimensionless, will With the first threshold Compare, This is a preset parameter, taken as an empirical value of 0.15, dimensionless. It can also be adjusted by the implementer according to the specific implementation scenario. A value greater than 0.15 indicates that the conductivity has undergone a significant cumulative deflection beyond the normal range of the dry zone within the sliding window.

[0055] Polarization response time autocorrelation coefficient It is a dimensionless coefficient and is directly related to the second threshold. Compare, The preset parameter is an empirical value of 0.35, which is dimensionless. It can also be adjusted by the implementer based on the polarization response characteristics. When the value is less than 0.35, it indicates that the self-similarity of the polarization phase sequence before and after the time delay is 0.5 seconds, has dropped to a low level, and the ion distribution at the electrode-soil interface is in a short-term rapid reconstruction state.

[0056] when conditions and If both conditions are met simultaneously, a humidity softening event is determined to have occurred, and the event flag is set. Set to 1 and record the current time. ,otherwise Keeping the value at 0 and continuing to slide the window for the next monitoring cycle, this event indicates that the current soil condition simultaneously exhibits two characteristics strongly correlated with abrupt changes in water content: a significant jump in conductivity and a short-term decorrelation of the polarization phase.

[0057] Step S4: When a humidity softening event occurs, retrieve the historical working condition case library and calculate the target excavation depth by weighting the inverse of the historical failure rate.

[0058] When the event flag When the value is 1, it indicates that the cumulative deflection gradient of conductivity has exceeded the first threshold and the autocorrelation coefficient of polarization phase lag has fallen below the second threshold. Both soil conductivity and interface polarization state show instantaneous jumps strongly correlated with abrupt changes in humidity, indicating that the high humidity zone ahead will lead to a significant increase in fruit breakage rate.

[0059] However, different site conditions, electrode burial status, and mechanical wear levels mean that the conductivity measurement and the optimal excavation depth are not a fixed functional relationship. If the current impedance value is directly substituted into the preset mapping calculation depth compensation amount, it is easy to over-compensate or under-compensate. Therefore, we turn to the historical case library to retrieve operation records with similar impedance characteristics and low failure rate, and synthesize the target depth compensation reference value required at present through statistical weighting.

[0060] exist Immediately after setting, extract the conductivity measurement of the latest sampling point from the sliding window. and polarization response index , forming the query vector ,in, The unit is ohms. Dimensionless, then for each record in the case library, the feature vector Similarity is calculated one by one. When the number of records in the case library is less than 3, Mahalanobis distance cannot be calculated, so Euclidean distance is used instead.

[0061] When the number of records in the case library reaches 3 or more, Mahalanobis distance is used:

[0062] In the formula, Given the covariance matrix of all eigenvectors in the case library, the number of records in the library must be at least 3; otherwise... Irreversible for The inverse matrix, This is a dimensionless distance value.

[0063] All After sorting in ascending order, extract the first... Each record is considered as the most similar alternative operating condition to the current operating condition. When the number of records in the case library is greater than or equal to 3 but less than M, M is taken as the actual number of records.

[0064] For the selected For each of the similar working conditions, extract their respective excavation depth setpoints. The unit is centimeters, representing the actual excavation depth used during the operation under that historical condition, and the corresponding historical breakthrough rate. , dimensionless, decimal representation.

[0065] The historical failure rate plus a preset non-zero small constant is used to calculate the failure rate. The reciprocal of the last digit is used as the weighting coefficient:

[0066] in, The empirical value is 0.01, which is dimensionless, and is used to prevent division by zero overflow when the historical failure rate of a record is zero.

[0067] All weighting coefficients are used for... Normalized weighted synthesis is performed to obtain the target mining depth. :

[0068] It should be noted that the actual fruit breakage rate recorded in the historical operation is used here, not the fruit breakage rate that has not yet been obtained.

[0069] The lower the historical failure rate, the better the effect of the current operation, and the greater its weight, thus... It tends to use depth settings that have historically yielded better control results.

[0070] Since the weighting coefficient is inversely related to the fruit breakage rate, the lower the fruit breakage rate, the greater the weight of the condition record. The final synthesized result... A depth setting that tends to yield lower fruit breakage rates in the past.

[0071] Should The desired depth will be compared with the current actual depth to obtain a depth correction instruction.

[0072] As soil texture, moisture baseline, and electrode surface condition gradually drift during continuous harvesting, the distribution of eigenvectors in the case library changes accordingly. If the covariance matrix... If this remains unchanged over a long period, the scale correction upon which the Mahalanobis distance relies will deviate from the true discrete structure of the current working condition, leading to biases in the retrieval of similar working conditions. Therefore, for every cumulative addition to the case database... After each record is completed, the entire historical feature vector in the database is used to recalculate the result. and To maintain adaptive accuracy in retrieval, Take 50 experience points, or the implementer can adjust the number of cases to be added to the database according to the speed at which cases are added.

[0073] The total number of records in the case library is set to a maximum of 2000. Once the maximum is reached, the oldest 200 records are deleted using a first-in, first-out (FIFO) strategy to ensure controllable memory space and reflect recent working conditions. During case library initialization, 10 typical dry soil working condition records can be pre-set, with a conductivity measure of 12 ohms, a polarization response index of 0.6, a digging depth of 8 cm, and a breakage rate of 0.02, to avoid the lack of available data during the cold start phase. The maximum number of records, the number of records to be deleted, and the parameter settings during cold start initialization can be configured by the implementer according to the specific implementation scenario.

[0074] Step S5: Output depth correction command in advance based on hydraulic hysteresis characteristics and harvester travel speed.

[0075] After reaching the target excavation depth After calculation, if the case library is insufficient, the default depth compensation value will be used, i.e. Centimeters, the onboard controller obtains the actual digging depth of the current blade from the displacement sensor of the excavator's hydraulic cylinder. The unit is centimeters, and the real-time travel speed of the harvester is obtained from the vehicle speed sensor. The unit is meters per second.

[0076] Meanwhile, the parameters of the first-order inertial plus pure hysteresis model were obtained offline by applying a step voltage to the hydraulic cylinder and recording the piston displacement response.

[0077] The identification method is as follows: With the harvester stationary, a step signal with an amplitude of 30% of the rated voltage is applied to the proportional valve of the hydraulic cylinder. Data from the piston displacement sensor is collected simultaneously, and the time from the issuance of the command to the displacement reaching 63.2% of the steady-state value is recorded as the inertial time constant. The time from the issuance of the command to the start of a significant change in displacement is recorded as the pure time delay. Of these, a significant change occurs exceeding 5% of the steady-state value. In this embodiment, Take 0.22 seconds of experience points. The empirical value is 0.18 seconds, and the unit is seconds. Each harvester should independently identify and store this set of parameters before leaving the factory.

[0078] Because there is a time lag between the hydraulic cylinder receiving the command voltage and the piston actually moving. The pure delay, and the displacement establishment process has first-order inertial characteristics, if at the event determination time If a depth correction command is issued directly, the depth change of the shovel blade will lag behind the arrival time of the leading edge of the soil moisture softening event. When the harvester is traveling at a normal operating speed, this lag causes the shovel blade to start adjusting only after entering the high humidity zone, which rapidly increases the fruit breakage rate in the high humidity zone.

[0079] Therefore, it is necessary to dynamically calculate the lead time based on the hysteresis characteristics and the harvester's travel speed, so that the command is issued in advance to offset the delay in hydraulic response.

[0080] The horizontal distance between the electrode array mounting point and the digging shovel is fixed as follows: Meters, the time required from detecting a moisture softening event to the blade reaching the front of the wet zone is The installation position of the electrode array can be adjusted by the implementer according to the specific implementation scenario.

[0081] Based on the conventional tuning of the step response of a first-order inertial plus pure time-delay system, the response lag time of the hydraulic actuator itself is... The inertial time constant and pure delay time Based on the offline step response experiment, the required instruction lead time is:

[0082] when At that time, no extra advance is needed; the instruction is issued at the normal time. In this case, the instructions need to be given in advance. issue.

[0083] If the harvester speed If it cannot be obtained, then the default is used. Second.

[0084] After obtaining the target depth compensation value With current depth Then, calculate the depth change. The unit is centimeters. A positive value indicates that the digging depth needs to be increased, while a negative value indicates that the digging depth needs to be decreased.

[0085] Because the hydraulic cylinder experiences opposite load forces when the blade rises and falls, and the pressure-bearing areas of the rod-side and rodless sides are different, there are differences in the response speed and overshoot tendency when reaching steady state in the two directions. If the command is too large when falling, it can easily cause the blade to penetrate the soil excessively, resulting in hydraulic shock and peanut damage. When rising, a faster retraction action is required to quickly reduce resistance and the risk of peanut breakage.

[0086] Therefore, a swing commutation coefficient is introduced when constructing depth correction commands. ,when When it is necessary to raise the blade, We take an empirical value of 1.2, which is dimensionless, to increase the contraction step size; when When the blade needs to be lowered, An empirical value of 0.8, dimensionless, is used to suppress the descent step size and mitigate the impact; when At that time, no new instructions will be issued.

[0087] Furthermore, differences in soil firmness, electrode wear, and harvester hydraulic system characteristics across different fields can lead to inconsistent actual digging depth adjustment effects from the same depth change, making it impossible to maintain a low fruit breakage rate under various working conditions with a fixed depth correction step.

[0088] Therefore, an adaptive compensation coefficient is introduced. The depth command step size is scaled overall; this coefficient is dimensionless and is applied during the system's initial startup. The empirical value of 1.0 is used, which is dimensionless and will be dynamically updated in subsequent steps based on online fruit breakage rate feedback. The latest value is used directly in this calculation. value.

[0089] Multiplying the above quantities together, we obtain the step size of the depth correction command in step form. The unit is centimeters. The symbol determines the direction of the hydraulic cylinder's movement.

[0090] In addition, it is necessary to verify whether the excavation depth after execution is within the safe range: set a minimum excavation depth. The maximum digging depth is 5 cm to avoid the shovel blade breaking through the soil. The blade should be 15 cm long to avoid damaging the fruit stem by cutting too deep.

[0091] like Then let .

[0092] like Then let .

[0093] Immediately after the event is determined, a step signal will be used to... The command terminal superimposed on the hydraulic cylinder position control loop, through the design of advance time, enables the actual displacement response of the hydraulic cylinder to complete most of the adjustment when the blade reaches the leading edge of the wet zone, thereby compensating for the action delay caused by pure hysteresis and inertial transition.

[0094] Step S6: Update the adaptive compensation coefficient based on real-time fruit breakage rate feedback.

[0095] A first-stage piezoelectric thin-film impact sensor array is installed at the bottom of the conveying channel 0.2 meters behind the excavator shovel exit. This array is used to detect fruit shell breakage caused by improper excavation depth. The sensor installation position can be adjusted by the implementer according to the specific implementation scenario.

[0096] Each impact pulse counts the excavation damage. Add 1: A photoelectric through-beam sensor is installed 0.1 meters behind the sensor in the same channel to measure the total number of particles passing through. .

[0097] Using 0.5 meters as the basic travel unit, the damage density of each 0.5-meter segment is calculated. .

[0098] To obtain a stable estimate of the breakage rate, the breakage counts and total number of seeds in four consecutive 0.5-meter segments (a total distance of 2 meters) are summed to obtain the breakage rate estimate. Estimated fruit breakage rate for each travel segment :

[0099] In the formula, Dimensionless, expressed as a decimal.

[0100] Due to an approximately 1.2-second transmission delay from excavation to damage detection, the depth command needs to be time-aligned with the damage pulse arriving 1.2 seconds later. No adaptive update is performed immediately after startup, before the first 2-meter travel is completed. The travel length is an empirical value that can be adjusted by the implementer based on the specific implementation scenario. This represents the real-time fruit breakage rate.

[0101] During continuous operation of the hydraulic cylinder, internal leakage of the piston seal, changes in the soil entry angle caused by blade wear, and non-uniform distribution of soil compaction can cause the actual blade displacement corresponding to the same depth correction command step size to drift, resulting in a change in the fixed adaptive compensation coefficient. Under these conditions, the fruit breakage rate shows a trend of increasing in the subsequent travel segment. At this point, information on the adjustment effects of adjacent segments can be used to... Perform sliding gradient descent correction to maintain a low fruit breakage rate.

[0102] When at least two consecutive fruit breakage rate estimates have been obtained and And satisfy When the adaptive compensation coefficient is updated, it is done using the following formula:

[0103] In the formula, For travel segment The compensation coefficient used internally is dimensionless and has an initial value of 1.0. For travel segment The compensation coefficient used internally is dimensionless. and These are the estimated fruit breakage rates for the corresponding travel segments, dimensionless; To learn the step size, an empirical value of 0.15 is used. This value is dimensionless and can be adjusted by the implementer according to the convergence speed requirements. As a stabilizing factor, an empirical value of 0.01 is taken. It is dimensionless to prevent the denominator from being too small. Implementers can adjust it according to the specific implementation scenario.

[0104] After the update Limiting amplitude: If Then set it to 0.5, if Then set it to 2.0. When At that time, skip this update and keep .

[0105] Updated Store and overwrite the original compensation coefficients, and then obtain the deep correction instructions later. At that time, multiplication factor All adopt the latest compensation coefficients, so that the step size of the depth correction command is continuously adaptively scaled according to the online fruit breakage rate feedback. At the same time, the estimated fruit breakage rate and feature vector of the current travel segment can be selectively stored in the historical working condition case library according to the case library maintenance strategy.

[0106] This invention also discloses a peanut harvester condition detection and control system, including a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the peanut harvester condition detection and control method of this invention. The system also includes other components well-known to those skilled in the art, such as a communication bus and communication interface; their configurations and functions are known in the art and will not be described further here.

[0107] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make several improvements and substitutions without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for detecting and controlling the operating conditions of a peanut harvester, characterized in that, include: Multi-frequency impedance measurement of soil is performed. The conductivity measure is determined based on the real part of the impedance in the low-frequency band, and the polarization response index is determined based on the phase angle in the high-frequency band. This includes arranging multiple annular electrodes at equal intervals on a crossbeam perpendicular to the direction of travel below the front edge of the harvester's digging shovel. Impedance excitation and acquisition are performed sequentially at multiple frequencies of a preset logarithmic distribution sequence; The average real part of the impedance at multiple frequency points in the low-frequency band is taken as the conductivity measure, and the average phase angle at multiple frequency points in the high-frequency band is taken as the polarization response index. Within a sliding window, the mean of the absolute differences between adjacent sampling points of the conductivity measure and the autocorrelation coefficient of the polarization response index under a preset lag time are calculated, including: a preset sliding window and a preset lag time; for the conductivity measure sequence within the sliding window, the arithmetic mean of the absolute differences between adjacent sampling points is calculated to obtain the mean of the absolute differences between adjacent sampling points of the conductivity measure; from the original sequence of the polarization response index, two sub-sequences are obtained by offsetting according to the preset lag time, and the Pearson correlation coefficient of the two sub-sequences is calculated to obtain the autocorrelation coefficient of the polarization response index under the preset lag time; When the mean value is greater than the first threshold and the autocorrelation coefficient is less than the second threshold, a humidity softening event is determined to have occurred. In response to the humidity softening event, similar operating conditions are retrieved from the historical operating condition case library, and the target depth compensation value is obtained by weighting the inverse of the fruit breakage rate. Based on the difference between the target depth compensation value and the current excavation depth, and combined with the hysteresis characteristics of the hydraulic actuator, a depth correction command is obtained and output in advance; Furthermore, the adaptive compensation coefficient is updated based on the online fruit breakage rate feedback, and the amplitude of the subsequent depth correction command is adjusted using the updated adaptive compensation coefficient.

2. The method for detecting and controlling the working condition of a peanut harvester according to claim 1, characterized in that, The condition that the mean is greater than a first threshold and the autocorrelation coefficient is less than a second threshold includes: The first threshold is the upper bound of the conductivity deflection gradient set based on the normalization of the impedance amplitude of the dry area; The second threshold is the lower bound of the phase angle lag autocorrelation coefficient.

3. The method for detecting and controlling the working condition of a peanut harvester according to claim 1, characterized in that, The step of retrieving similar working conditions from the historical working condition case database and calculating the target depth compensation value by weighting the result by the inverse of the failure rate includes: Each record in the historical working condition case library includes conductivity measurement, polarization response index, mining depth correction value, and fruit breaking rate. Calculate the Mahalanobis distance between the query vector composed of the current conductivity metric and polarization response index and the feature vectors of each record in the historical operating condition case library; determine the preset number of records with the smallest Mahalanobis distance as the similar operating conditions; The target depth compensation value is obtained by weighting the excavation depth correction value corresponding to the similar working conditions by using the reciprocal of the break rate plus a preset non-zero small constant as the weight.

4. The method for detecting and controlling the working condition of a peanut harvester according to claim 1, characterized in that, The step of obtaining a depth correction command and outputting it in advance based on the difference between the target depth compensation value and the current excavation depth, combined with the hysteresis characteristics of the hydraulic actuator, includes: Obtain the inertial time constant and pure time delay in the first-order inertial plus pure time delay model of the hydraulic actuator; The lead time is calculated as the pure lag time plus twice the inertial time constant. Based on the difference between the target depth compensation value and the current excavation depth, a step-type depth correction command is generated and issued ahead of the specified lead time to compensate for the lag of the hydraulic actuator.

5. The method for detecting and controlling the working condition of a peanut harvester according to claim 1, characterized in that, The step of updating the adaptive compensation coefficient based on online fruit breakage rate feedback, and adjusting the amplitude of subsequent depth correction commands using the updated adaptive compensation coefficient, includes: Obtain the estimated fruit breakage rate for the current travel segment and the estimated fruit breakage rate for the previous travel segment; Based on the difference between the current value and the previous value of the adaptive compensation coefficient, and combined with the change in the estimated fruit breakage rate, the sliding gradient descent method is used to calculate the compensation coefficient update amount and update the adaptive compensation coefficient.

6. The method for detecting and controlling the working condition of a peanut harvester according to claim 4, characterized in that, The depth correction command includes: Based on the comparison between the target depth compensation value and the current excavation depth, the swing reversal coefficient is set; When the target depth is less than the current depth and the excavation depth needs to be reduced, the swing reversal coefficient is set to be greater than 1. When the target depth is greater than the current depth and the excavation depth needs to be increased, the swing reversal coefficient is set to be less than 1, and the step-type depth correction command is multiplied by the swing reversal coefficient.

7. A peanut harvester operating condition detection and control system, characterized in that, include: The processor and memory, wherein the memory stores computer program instructions that, when executed by the processor, implement a peanut harvester condition detection and control method according to any one of claims 1 to 6.

Images