A method for measuring the droplet size of an ultrasonic nebulizer based on laser diffraction

Abstract

The present application relates to the technical field of agricultural plant protection spraying detection, in particular to a kind of ultrasonic sprayer mist particle size measurement method based on laser diffraction, comprising: obtaining the laser diffraction signal of multiple target measurement positions and overlapping measurement positions, collection time and nozzle swing angle data;According to the start and end point of the spray cycle determined by the nozzle swing angle, the laser diffraction signal is mapped to the corresponding cycle stage and the candidate equivalent signal is extracted;The candidate equivalent signal of the overlapping measurement position is carried out particle size inversion, and the target signal is screened according to the particle size distribution difference of different spray cycles;The target signal of different target measurement positions under the same cycle stage is spatially reorganized, and particle size inversion and splicing are carried out in combination with spatial coordinates, to generate the spatial mist particle size distribution of the corresponding cycle stage.The present application realizes the measurement of mist particle size spatial distribution under dynamic spraying condition by cycle matching and spatial reorganization of laser diffraction signal.

Description

Technical Field

[0001] This invention relates to the field of agricultural plant protection spray detection technology, specifically a method for measuring the droplet size of ultrasonic sprayers based on laser diffraction. Background Technology

[0002] As agricultural plant protection equipment develops towards higher efficiency, precision, and lower pollution, agricultural sprayers are increasingly being used for the atomization of liquids such as pesticides and fertilizers. Ultrasonic sprayers utilize high-frequency vibrations to form droplets from liquids, offering features such as stable atomization and adjustable droplet size. In actual spraying operations, droplet size and its distribution are correlated with pesticide deposition, coverage, and spraying effectiveness. Therefore, measuring droplet size during spraying provides a data foundation for adjusting sprayer operating parameters and evaluating spraying performance.

[0003] Currently, droplet size can be measured using methods such as image analysis, phase Doppler ultrasound, and laser diffraction. Among these, laser diffraction obtains the energy distribution of scattered light generated by the droplets to the incident laser and, combined with a corresponding scattering model, inverts the droplet size distribution, enabling non-contact measurement of spray droplet size.

[0004] In actual operation, agricultural sprayers allow the nozzles to oscillate according to operational needs, and the spraying platform can also be mobile, causing the nozzle posture and spraying stage to change over time at different spatial locations. For a spraying area with a certain spatial range, when laser diffraction signals are collected sequentially from multiple measurement locations, the measurement data at each location may correspond to different spraying cycles or different cycle stages. Therefore, when using measurement data from different locations to characterize the spatial droplet size distribution, it is also necessary to consider the correspondence between the spraying states at different acquisition times.

[0005] Furthermore, factors such as the liquid supply status, nozzle oscillation status, and operating platform movement status can all change during the spraying process, and the droplet state may also differ between different spraying cycles. Therefore, there is a need for further research on how to combine the spraying operation status to correspond and filter laser diffraction signals acquired at different times and spatial locations, and further form droplet size measurement results with spatial location correspondence.

[0006] In summary, how to achieve correlation measurement of droplet size at different spatial locations under dynamic spraying conditions is a technical problem that urgently needs to be solved in this field.

[0007] To address this, a method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction is proposed. Summary of the Invention

[0008] The purpose of this invention is to provide a method for measuring droplet size in an ultrasonic sprayer based on laser diffraction. This method achieves measurement of the spatial distribution of droplet size under dynamic spraying conditions by periodically matching and spatially recombining laser diffraction signals. The method includes: acquiring laser diffraction signals, acquisition times, and nozzle oscillation angle data from multiple target measurement locations and overlapping measurement locations; determining the start and end points of the spray cycle based on the nozzle oscillation angle; mapping the laser diffraction signals to the corresponding cycle stages and extracting candidate equivalent signals; performing droplet size inversion on the candidate equivalent signals at overlapping measurement locations; and filtering target signals based on differences in droplet size distribution across different spray cycles; spatially recombining the target signals from different target measurement locations within the same cycle stage; and combining spatial coordinates for droplet size inversion and stitching to generate the spatial droplet size distribution for the corresponding cycle stage.

[0009] To achieve the above objectives, the present invention provides the following technical solution: A method for measuring droplet size in an ultrasonic sprayer based on laser diffraction, comprising: The spatial coordinates of multiple target measurement positions and overlapping measurement positions within the spray area are obtained. The laser diffraction signals of the target measurement positions and overlapping measurement positions and the corresponding acquisition time are collected. The nozzle swing angle data of the ultrasonic sprayer in multiple spray cycles are also obtained. The start and end points of multiple spray cycles are determined based on the nozzle swing angle data. The laser diffraction signals in each spray cycle are mapped to the corresponding cycle stage using the acquisition time. The laser diffraction signals in the same cycle stage in different spray cycles are extracted as candidate equivalent signals. A non-negative least squares inversion algorithm is used to invert the candidate equivalent signals corresponding to the overlapping measurement locations to obtain the overlapping particle size distribution vector. The difference between the overlapping particle size distribution vectors corresponding to different spray cycles is calculated, and the candidate equivalent signals with a difference value less than a set threshold are selected as the target signals. The target signals corresponding to the same cycle stage and located at each target measurement location are combined into spatial reconstructed data. The non-negative least squares inversion algorithm is used to invert the spatially reconstructed data. The inversion results are stitched together using the spatial coordinates of the target measurement location to generate the spatial droplet size distribution of the ultrasonic sprayer in the corresponding period stage.

[0010] Preferably, the process of acquiring the spatial coordinates of multiple target measurement positions within the spray area, as well as overlapping measurement positions, collecting laser diffraction signals of the target measurement positions and overlapping measurement positions and the corresponding acquisition times, and acquiring nozzle oscillation angle data of the ultrasonic sprayer in multiple spray cycles includes: The coordinates of the target measurement positions are extracted in a pre-built three-dimensional coordinate system as the spatial coordinates; a fixed diffraction optical path is deployed at the overlapping measurement positions, and the moving diffraction optical path is controlled to sequentially reach each of the target measurement positions; light energy distribution data is synchronously collected through the fixed diffraction optical path and the moving diffraction optical path as the laser diffraction signal, and the timestamp corresponding to the light energy distribution data is extracted as the acquisition time; the Mie scattering response matrix is ​​calculated based on the optical parameters of the fixed diffraction optical path and the moving diffraction optical path; the real-time angle sequence of the ultrasonic sprayer shaft is continuously collected using an angle sensor, and the real-time angle sequence is used as the nozzle swing angle data.

[0011] Preferably, the process of extracting laser diffraction signals in the same period phase during different spraying cycles as candidate equivalent signals includes: acquiring the driving state data, liquid supply state data, nozzle attitude state data, and spraying platform motion state data of the ultrasonic sprayer; inputting the driving state data, liquid supply state data, and nozzle swing angle data into a state event recognition model; and extracting the timestamps of spray start-up events, liquid supply state transition events, and nozzle swing reversal events as start and end points; inputting the driving state data, liquid supply state data, nozzle attitude state data, spraying platform motion state data, and nozzle swing angle data into a multi-source data fusion model; and extracting the liquid supply flow rate using the liquid supply state data, with the liquid supply flow rate being greater than... A preset flow rate threshold is used as a data validity constraint. Data that meets the data validity constraint is sampled, aligned, and dimensionless, and a multi-dimensional state feature matrix corresponding to the spray cycle is output as the spray state sequence. Using the acquisition time, each laser diffraction signal is bound to the corresponding node of the spray state sequence to generate time-series mapping signal data. The spray state sequences of different spray cycles are input into the time-series alignment model, mapped to the unified periodic process coordinates based on the start and end points, and divided at fixed intervals to obtain multiple phase intervals as periodic stages. Using the time-series mapping signal data, laser diffraction signals that are in the same periodic stage and whose corresponding spray state sequences have the same state change order are extracted as candidate equivalent signals.

[0012] Preferably, the process of obtaining the overlapping particle size distribution vector includes: The background light reference signal is subtracted from the signals collected by the fixed diffraction path and the moving diffraction path in the candidate equivalent signals to obtain the net scattered light intensity vector of the fixed path and the net scattered light intensity vector of the moving path. The net scattered light intensity vector of the fixed path and the net scattered light intensity vector of the moving path are divided by their respective sampling volumes and then vertically concatenated to generate a joint light intensity vector. The submatrices corresponding to the fixed diffraction path and the moving diffraction path in the Mie scattering response matrix are vertically concatenated to generate a joint scattering response matrix. Under the constraint of the smoothness regularization factor, the non-negative least squares inversion algorithm is used to invert the joint light intensity vector using the joint scattering response matrix to output the overlapping particle size distribution vector.

[0013] Preferably, the process of selecting candidate equivalent signals with a difference value less than a set threshold as the target signal includes: For each candidate spray cycle containing a candidate equivalent signal, calculate the sum of the Euclidean distances between the overlapping particle size distribution vector corresponding to that candidate spray cycle and the overlapping particle size distribution vectors corresponding to all other candidate spray cycles containing the candidate equivalent signal. Use the sum of the Euclidean distances as the difference value corresponding to that candidate spray cycle. Summarize the difference values ​​corresponding to all candidate spray cycles, calculate the arithmetic mean of all difference values, and use the arithmetic mean as a set threshold. Extract candidate spray cycles with difference values ​​less than the set threshold as selected spray cycles, and determine the candidate equivalent signal corresponding to the selected spray cycle as the target signal.

[0014] Preferably, the process of combining spatially reconstructed data includes: Acquire nozzle attitude state data and spray operation platform motion state data when collecting the laser diffraction signal; calculate the three-dimensional spatial coordinates and spray attitude angle of each target measurement position based on the nozzle attitude state data and spray operation platform motion state data; map the target signal corresponding to each target measurement position to a three-dimensional spatial grid using the three-dimensional spatial coordinates and the spray attitude angle to obtain node signal data; perform grid interpolation and spatial matrix splicing on each node signal data to generate the spatial reconstruction data.

[0015] Preferably, the process of generating the spatial droplet size distribution of the ultrasonic sprayer in the corresponding periodic stage includes: Scattered light intensity data of each target measurement location is extracted from the spatial reconstructed data. A non-negative least squares inversion algorithm is used to invert the scattered light intensity data of each target measurement location to obtain the particle size distribution vector of each target measurement location. The spatial coordinates of each target measurement location are obtained, and each particle size distribution vector is mapped to the corresponding spatial coordinates and spliced ​​according to the spatial position relationship to generate a global spatial particle size distribution matrix. The time interval is determined according to the start and end points and the corresponding periodic stage. The particle size data located in the time interval in the global spatial particle size distribution matrix is ​​extracted to generate the spatial droplet particle size distribution of the ultrasonic sprayer in the corresponding periodic stage.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention combines the driving state, liquid supply state, nozzle attitude state, spraying platform motion state, and nozzle swing angle of the ultrasonic sprayer to uniformly process state data from different sources. It determines the spray cycle and cycle stages based on events such as spray initiation, liquid supply state transition, and nozzle swing reversal. Furthermore, it maps laser diffraction signals to spray state sequences according to acquisition time and performs unified process mapping on different spray cycles, enabling comparison of spray cycles of different durations on the same cycle scale. This allows for the selection of candidate equivalent signals within the same cycle stage with consistent state change order, providing a state correspondence basis for subsequent comparison of measurement data from different spray cycles and particle size distribution calculations.

[0017] 2. This invention reduces the influence of ambient light and detector baseline on scattered light intensity data by separately subtracting the background light reference signals of the fixed and moving diffraction optical paths. Furthermore, it normalizes the net scattered light intensity of the two optical paths using sampling volume, providing a unified basis for comparison of data acquired from different optical paths. The invention further constructs the light intensity vectors of the two optical paths and their corresponding Mie scattering response sub-matrices, and performs particle size inversion using a non-negative least squares algorithm under smoothness regularization constraints. This ensures the inversion results satisfy the non-negative particle size distribution constraint while reducing abnormal fluctuations in adjacent particle size intervals, thereby forming an overlapping particle size distribution vector with a unified data format. This provides a data basis for comparing particle size distributions between different spray cycles.

[0018] 3. This invention combines the nozzle attitude state during laser diffraction signal acquisition with the motion state of the spraying platform to uniformly determine the spatial coordinates and spray attitude of different target measurement positions. This allows target signals obtained under different acquisition times, nozzle attitudes, and platform positions to be converted into a unified three-dimensional spatial coordinate system. Furthermore, based on the three-dimensional spatial position coordinates and spray attitude angles, the target signals are mapped to corresponding spatial grids. Spatial reconstructed data is then formed through grid interpolation and spatial matrix splicing, thereby establishing the correspondence between each target signal and its spatial position. This enables laser diffraction data acquired at different positions to be organized according to spatial position, providing a structured data foundation for subsequent particle size inversion and spatial droplet size distribution generation. Attached Figure Description

[0019] Figure 1 A schematic flowchart of a method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction, provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the process of extracting laser diffraction signals in the same period stage of different spray cycles into candidate equivalent signals, provided for an embodiment of the present invention; Figure 3 This is a schematic diagram of the process for obtaining the overlapping particle size distribution vector provided in an embodiment of the present invention. Detailed Implementation

[0020] 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 embodiments of the present invention, and not all embodiments. 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.

[0021] Please see Figures 1 to 3 This invention provides a method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction, the technical solution of which is as follows: Example 1: A method for measuring droplet size in an ultrasonic sprayer based on laser diffraction, such as... Figure 1 As shown, it includes: The spatial coordinates of multiple target measurement positions and overlapping measurement positions within the spray area are obtained. The laser diffraction signals of the target measurement positions and overlapping measurement positions and the corresponding acquisition time are collected. The nozzle swing angle data of the ultrasonic sprayer in multiple spray cycles are also obtained. The start and end points of multiple spray cycles are determined based on the nozzle swing angle data. The laser diffraction signals in each spray cycle are mapped to the corresponding cycle stage using the acquisition time. The laser diffraction signals in the same cycle stage in different spray cycles are extracted as candidate equivalent signals. A non-negative least squares inversion algorithm is used to invert the candidate equivalent signals corresponding to the overlapping measurement locations to obtain the overlapping particle size distribution vector. The difference between the overlapping particle size distribution vectors corresponding to different spray cycles is calculated, and the candidate equivalent signals with a difference value less than a set threshold are selected as the target signals. The target signals corresponding to the same cycle stage and located at each target measurement location are combined into spatial reconstructed data. The non-negative least squares inversion algorithm is used to invert the spatially reconstructed data. The inversion results are stitched together using the spatial coordinates of the target measurement location to generate the spatial droplet size distribution of the ultrasonic sprayer in the corresponding period stage.

[0022] Furthermore, the process of acquiring the spatial coordinates of multiple target measurement positions within the spray area, as well as overlapping measurement positions, collecting laser diffraction signals of the target measurement positions and overlapping measurement positions and their corresponding acquisition times, and acquiring nozzle oscillation angle data of the ultrasonic sprayer in multiple spray cycles includes: The coordinates of the target measurement positions are extracted in a pre-built three-dimensional coordinate system as the spatial coordinates; a fixed diffraction optical path is deployed at the overlapping measurement positions, and the moving diffraction optical path is controlled to sequentially reach each of the target measurement positions; light energy distribution data is synchronously collected through the fixed diffraction optical path and the moving diffraction optical path as the laser diffraction signal, and the timestamp corresponding to the light energy distribution data is extracted as the acquisition time; the Mie scattering response matrix is ​​calculated based on the optical parameters of the fixed diffraction optical path and the moving diffraction optical path; the real-time angle sequence of the ultrasonic sprayer shaft is continuously collected using an angle sensor, and the real-time angle sequence is used as the nozzle swing angle data.

[0023] A three-dimensional Cartesian coordinate system is established with the center of the ultrasonic sprayer nozzle as the origin. The movement direction of the spraying platform is set as the X-axis, the horizontal direction perpendicular to the X-axis is set as the Y-axis, and the vertical direction is set as the Z-axis. Multiple measurement nodes are set at 100 mm intervals within the spraying area. The coordinate values ​​of each measurement node in the three-dimensional Cartesian coordinate system are extracted, and the measurement node is determined as the target measurement position. A measurement node located in the middle of the spraying area is selected as the overlapping measurement position.

[0024] A fixed diffraction optical path is set at the overlapping measurement position, and a moving diffraction optical path is mounted on a three-axis moving mechanism. The three-axis moving mechanism controls the moving diffraction optical path to sequentially reach each target measurement position according to its spatial coordinates. After reaching the corresponding target measurement position, the moving diffraction optical path stops moving and acquires the laser diffraction signal at that position. While the moving diffraction optical path sequentially acquires the laser diffraction signals at each target measurement position, the fixed diffraction optical path continuously acquires the laser diffraction signals at the overlapping measurement position and calculates the sliding average value of the net scattered light intensity vector of the fixed optical path in real time as a reference state. When the relative change between the current sliding average value and the light intensity reference value during the last dual-path synchronous calibration exceeds 5%, or when the number of target measurement positions continuously acquired by the moving diffraction optical path reaches 5, the traversal of target measurement positions is paused, and the moving diffraction optical path is controlled to return to the overlapping measurement position, allowing both the fixed and moving diffraction optical paths to simultaneously acquire the laser diffraction signals at the overlapping measurement position to update the calibration state. After completing synchronous acquisition, the moving diffraction optical path is controlled to continue to the next unacquired target measurement position.

[0025] Both the fixed and moving diffraction optical paths are equipped with a continuous-wave laser emitter, a collimation assembly, a receiving lens, a ring photodetector, a multi-channel programmable gain amplifier array, and a signal acquisition unit. The laser emitted by the continuous-wave laser emitter is collimated and passes through the corresponding measurement position. The forward scattered light generated by the droplets is focused by the receiving lens and received by the ring photodetector.

[0026] Considering the significant differences in light intensity at different scattering angles among droplets of different sizes ranging from 1 μm to 500 μm, gain parameters were configured for each detection channel. Specifically, the paraxial channel, corresponding to the region of large particles and small-angle scattering, used a lower gain to prevent strong scattering signals from entering the saturation region; while the peripheral channels, corresponding to the region of small particles and large-angle scattering, used a higher gain to enhance weak scattering signals. After gain adjustment, each channel outputs its corresponding light energy value, which, according to the channel position order, forms the light energy distribution data of a single sample, serving as the laser diffraction signal.

[0027] When spray droplets pass through the measurement area at high speed, the instantaneous scattering signal is prone to fluctuations. To mitigate the impact of these fluctuations on the measurement results, the signal acquisition unit performs internal sampling at a frequency of 10 kHz and uses a 10 ms integration time window to average 100 sets of transient light energy data continuously acquired within the window. After processing, both the fixed and moving diffraction optical paths output valid laser diffraction signals at a frequency of 100 Hz, and both are initiated by the same synchronous trigger signal. Simultaneously, the time information from a unified timing module is read when each set of valid signals is output, and this time value is used as the acquisition time of the corresponding laser diffraction signal.

[0028] Based on the laser wavelengths of the fixed and moving diffraction paths, the focal length of the receiving lens combination, the parameters of each forward detection channel of the ring photodetector, the parameters of the lateral large-angle detector, and the photoelectric response parameters, a corresponding scattering response relationship is established. To expand the scattering angle detection range, in addition to the ring photodetector for receiving the forward small-angle scattered light from large-diameter droplets, a lateral large-angle detector is also installed on the side of the receiving lens to receive the larger-angle scattered light generated by small-diameter droplets.

[0029] The droplet size range to be measured was set to 1 μm to 500 μm, and divided into 60 droplet size intervals at logarithmic intervals. Using the representative droplet size of each interval as the calculation parameter, and combining the laser wavelength, the complex refractive index of the drug solution (including the real and imaginary absorption parts), the polarization state of the incident light, and the spectral transmittance of the receiving lens and the protective window, a double numerical integration was performed on the solid angle range corresponding to each detection channel. The scattering response coefficients of each detection channel for different droplet size intervals were obtained according to Mie scattering theory, and the matrix version number was recorded and archived. This resulted in scattering response sub-matrices for fixed diffraction paths, each with dimensions of 32 × 60. Scattering response submatrix of moving diffraction path Its elemental physical unit is The Mie scattering response matrix required for subsequent inversion is composed of two sub-matrices.

[0030] Before performing the scattering response calculation, the intensity of the transmitted light after the laser passes through the droplet field is measured using the central channel of the ring photodetector. and the transmitted light intensity under droplet-free conditions By comparison, the transmittance is obtained. .when( When the current droplet field is considered to satisfy the single scattering approximation condition, the response matrix can be directly established using Mie scattering theory; when ( If the current spray concentration or the measured optical path is not suitable for inversion under the above conditions, it will suggest reducing the optical path or reducing the spray flow rate and then measuring again.

[0031] An angle sensor is installed at the shaft of the ultrasonic sprayer, and the midpoint of the nozzle's swing stroke is defined as 0°. The angle sensor continuously outputs the real-time angle of the shaft at a frequency of 200Hz, and a unified timing module synchronously records the acquisition time of each angle value. The real-time angle values ​​are arranged in chronological order to form a sequence of nozzle swing angles.

[0032] The laser diffraction signal and the nozzle oscillation angle data use the same timing reference. Based on the timestamp, the 100 Hz output laser diffraction signal can be correlated with the 200 Hz acquired nozzle angle data, thereby determining the oscillation position and oscillation stage of the nozzle at the time of each diffraction signal acquisition.

[0033] Within a unified three-dimensional coordinate system, the target measurement positions corresponding to the fixed and moving optical paths, as well as the overlapping measurement positions of the two optical paths, are pre-determined to establish a correspondence between the acquired signals and spatial positions. Since the two diffraction optical paths share a synchronous trigger signal and a unified timing module, the light energy distribution data obtained from different spatial positions have comparable time signatures. Combined with the scattering response sub-matrices independently established for each optical path, this can be used for subsequent joint inversion of multi-position data. The time correspondence between the nozzle angle sequence and the laser diffraction signal is used to determine the spray cycle and cycle stage to which each sampled data belongs, and provides basic data for subsequent screening of candidate equivalent signals and reconstructing the spatial droplet size distribution.

[0034] Furthermore, the process of extracting laser diffraction signals from the same period stage in different spray cycles as candidate equivalent signals includes, for example... Figure 2 As shown: Acquire the drive status data, liquid supply status data, nozzle attitude status data, and spray platform motion status data of the ultrasonic sprayer. Input the drive status data, liquid supply status data, and nozzle swing angle data into a status event recognition model to extract the timestamps of spray start events, liquid supply status transition events, and nozzle swing direction reversal events as start and end points. Input the drive status data, liquid supply status data, nozzle attitude status data, spray platform motion status data, and nozzle swing angle data into a multi-source data fusion model. Use the liquid supply status data to extract the liquid supply flow rate. A liquid supply flow rate greater than a preset flow rate threshold is used as a data validity constraint. Data meeting the threshold is then processed. Data with validity constraints is sampled, aligned, and dimensionless, outputting a multidimensional state feature matrix corresponding to the spray cycle as the spray state sequence. Using the acquisition time, each laser diffraction signal is bound to the corresponding node of the spray state sequence to generate time-series mapping signal data. Spray state sequences from different spray cycles are input into the time-series alignment model, mapped to a unified periodic process coordinate system based on start and end points, and divided at fixed intervals to obtain multiple phase intervals as periodic stages. Using the time-series mapping signal data, laser diffraction signals in the same periodic stage and corresponding spray state sequences with the same state change order are extracted as candidate equivalent signals.

[0035] The system acquires driving state data, liquid supply state data, nozzle attitude state data, and spray platform motion state data for the ultrasonic sprayer. Driving state data includes the driving current, driving voltage, and operating frequency of the ultrasonic transducer; liquid supply state data includes the start / stop status of the liquid supply pump, liquid flow rate, and liquid pressure; nozzle attitude state data includes the nozzle pitch and yaw angles; and spray platform motion state data includes platform position, moving speed, and moving acceleration. Each type of state data is recorded with a corresponding acquisition timestamp. The driving state data, liquid supply state data, and nozzle oscillation angle data are input into the state event recognition model. The ultrasonic sprayer operates at frequencies from 20 kHz to 40 kHz. The relationship between the corresponding droplet size and ultrasonic frequency is given by the Lang formula. A theoretical droplet size estimate corresponding to the ultrasonic frequency is added as an auxiliary feature channel to the input features of the state event recognition model, enabling the model to utilize the physical correlation between ultrasonic frequency and atomization state during event recognition. The state event recognition model consists of a one-dimensional convolutional layer, a bidirectional long short-term memory network layer, and a fully connected classification layer connected sequentially. The system consists of seven input channels for the one-dimensional convolutional layer, corresponding to seven input features: driving current, driving voltage, operating frequency, theoretical droplet size estimate, liquid supply pump start / stop status, liquid supply flow rate, and nozzle oscillation angle. The one-dimensional convolutional layer has 32 convolutional kernels with a kernel size of 5 and a stride of 1, and uses a linear rectified function as the activation function. The bidirectional long short-term memory network layer has 64 hidden units and outputs a bidirectionally concatenated 128-dimensional feature vector. The fully connected classification layer has four output nodes, corresponding to spray start-up events, liquid supply state transition events, nozzle oscillation reversal events, and non-event states, respectively, and uses a softmax function as the activation function.

[0036] The training data includes 200 hours of operating data collected from 5 different batches of ultrasonic sprayers, covering 3 typical viscosity solutions (pure water, conventional pesticide aqueous solutions, and high viscosity suspensions), with the sliding window length set to 1 second. Each event label is defined using an independent protocol of "rule-based initial screening + manual physical truth value verification" as follows: First, the first sampling moment within a time window where the driving current is continuously higher than the no-load current threshold for a duration of 0.2 seconds is marked as the initial screening point for the spray start-up event. This is then manually corrected using the physical time of the first droplet detachment from the nozzle in a simultaneously captured high-speed video, serving as the independent truth value label for the spray start-up event. The first sampling moment within a time window where the liquid supply flow rate exceeds 5 ml / min for a duration of 0.2 seconds is marked as the initial screening point for the liquid supply state transition event. This is then manually verified using the sudden jump signal from the pipeline micro-pressure sensor. The sampling moment where the sign of the nozzle oscillation angular velocity changes and the absolute value of the angular velocity is less than 1 degree per second is marked as the initial screening point for the nozzle oscillation reversal event, calibrated using the limit switch signal of the oscillation mechanism. All other sampling moments outside the aforementioned time windows are marked as non-event states. Double-blind labeling is performed by two data labelers, achieving a Kappa consistency coefficient of over 0.92 to ensure that the training labels do not deviate from pure rule fitting.

[0037] The model was trained using the cross-entropy loss function and the Adam optimizer, with an initial learning rate of 0.001, a batch size of 32, and 50 training iterations. An early stopping mechanism was introduced, stopping training when the validation set loss did not decrease for 10 consecutive epochs to prevent overfitting. A global random seed of 42 was set to ensure reproducibility. After training, for real-time input data including drive current, drive voltage, operating frequency, pump start / stop status, flow rate, and nozzle oscillation angle, the system calculated the theoretical droplet size estimate online using the Lang formula based on the current operating frequency and the surface tension and density of the liquid. This estimate was used as the fourth feature channel, forming a 7-channel time series with the aforementioned six data types. The model outputs the predicted event category probability value for each sampling moment using a sliding window approach. Intervals where three or more consecutive sampling moments predict the same event category with a probability greater than 0.8 were defined as event occurrence intervals, and the timestamp of the first sampling moment within this interval was taken as the timestamp of the corresponding event. This model, designed to replace rule-defined event labels, provides robust event recognition even in situations where threshold boundaries are blurred or rules fail due to signal noise during actual spraying operations. To prevent data leakage, the training data was grouped and segmented according to equipment number and pesticide batch, divided into training, validation, and test sets in a 7:2:1 ratio. Validation on the test set showed that the precision rates for spray initiation, pesticide supply status transition, and nozzle reversal events reached 98.2%, 97.5%, and 99.1%, respectively; the recall rates were 97.8%, 96.9%, and 98.8%, respectively; the F1 scores were 98.0%, 97.2%, and 98.9%, respectively; and the root mean square error of event time was less than 15 milliseconds. Even under harsh operating conditions with artificially injected 15% random electromagnetic noise and in cross-domain tests with unseen pesticides, the model's F1 score remained above 95.3%, demonstrating a significant improvement over traditional single-threshold rules.

[0038] The trained state event recognition model is used to extract timestamps for each event. The timestamp of the spray initiation event is used as the starting point of the first spray cycle, and the timestamps of two adjacent nozzle oscillation reversal events in the same direction are used as the starting and ending points of subsequent spray cycles, respectively. The timestamps of the liquid supply state transition events are used to define the effective liquid supply time period within each spray cycle.

[0039] After the data on drive status, liquid supply status, nozzle attitude status, spray platform motion status, and nozzle swing angle are collected, the data from these different sources are processed uniformly. The processing mainly includes unifying the sampling time, removing invalid data, converting parameter scales, and combining status parameters.

[0040] Various status parameters are arranged in a pre-defined order, including drive current, operating frequency, drive voltage, liquid supply pump start / stop status, liquid supply flow rate, liquid supply pressure, nozzle swing angle, nozzle pitch angle, nozzle yaw angle, platform X-axis position, platform Y-axis position, platform Z-axis position, platform moving speed, and platform moving acceleration, totaling 14 items. During arrangement, parameters with direct physical relationships are placed in adjacent positions, such as drive current and operating frequency, liquid supply flow rate and liquid supply pressure, and nozzle swing angle and nozzle attitude angle, to preserve the correlation information between different status parameters during subsequent data processing.

[0041] Since the data sampling frequencies of each sensor are not completely consistent and there may be slight clock deviations between different hardware modules, a 200Hz sampling time series of the nozzle oscillation angle is used as the reference time axis. Data from each sensor is globally time-aligned based on the hardware timestamp of the unified timing module. For locally missing values ​​caused by communication packet loss, linear interpolation is performed using adjacent valid sample values ​​before resampling. Let the original sampling frequency of the remaining k-th type of state data be... .when At Hz, linear interpolation is performed based on adjacent original sampled values ​​to obtain the state value corresponding to the 200Hz time node; when To prevent spectral aliasing caused by direct decimation at Hz, a fourth-order Butterworth digital low-pass anti-aliasing filter with a cutoff frequency of 100Hz (i.e., the target Nyquist frequency) is first applied to the original time series. To eliminate the phase group delay introduced by the filter, forward and backward bidirectional filtering is employed. After filtering, the time series is then processed with each 200Hz reference time node as the center, and the time series before and after it is processed. seconds (where the target frequency) The filtered sampled values ​​within a range of Hz (i.e., 2.5 milliseconds before and after) are arithmetically averaged and downsampled, and the result is used as the state value at that time node. Through the above processing, various state data are transformed onto the same sampling time axis, effectively ensuring that the event delay error and phase error after multi-source heterogeneous signal alignment are strictly controlled within the limit of a single target sampling period (5 milliseconds).

[0042] After time alignment, scaling is performed based on the statistical distribution of each continuous state parameter in the training data. Specifically, the mean and standard deviation of each parameter are calculated, and the original values ​​are transformed using the Z-score method to reduce the influence of different physical dimensions and numerical ranges on subsequent calculations. The 14 standardized parameters constitute the fused state data. When performing spray cycle matching, five parameters that reflect the spray working state and motion process—drive current, liquid supply flow rate, nozzle oscillation angle, nozzle pitch angle, and platform movement speed—are selected to form a state feature vector for time series analysis.

[0043] The liquid supply flow rate is also used to determine whether the current sampling time is in an effective spray state. Based on the stable spray test results of the liquid supply pump, 5 mL / min is set as the judgment threshold; when the flow rate exceeds this value, the corresponding sampling data participates in subsequent processing, and data that does not reach this value is not included in the effective spray dataset.

[0044] Based on the sampling time sequence, the state feature vectors retained within the same spray cycle are arranged sequentially to form the state feature matrix corresponding to that spray cycle. This matrix records the changes in the driving, liquid supply, nozzle attitude, and platform motion states over time during the spraying process and serves as the data input for time-series alignment of subsequent spray cycles.

[0045] For each set of laser diffraction signals, the sampling node with the closest time in the spray state sequence is searched according to its acquisition time, and a time-series correspondence between the two is established to form time-series mapped signal data. If the time interval between the signal acquisition time and two adjacent sampling nodes is equal, the earlier sampling node is selected to correspond to the laser diffraction signal.

[0046] A dynamic time warping algorithm is used for time alignment between different spray cycles. During calculation, standardized drive current, liquid flow rate, nozzle oscillation angle, nozzle pitch angle, and platform movement speed are used as state features. The local distance between alignment points is represented by the Euclidean distance of the corresponding multidimensional state feature vectors. To limit unreasonable time axis stretching, the path search of the dynamic time warping algorithm uses a Sakoe-Chiba constraint window, with a window width of 10% of the length of the alignment sequence.

[0047] After calculating the cumulative distance, the optimal regularization path is traced in reverse to obtain the correspondence between the sampling nodes of each spray cycle to be aligned and the reference cycle. The time axis of the reference cycle is normalized to the range of 0 to 1, and based on the corresponding position of each sampling node in the regularization path on the reference cycle, the sampling nodes of the cycle to be aligned are assigned uniform cycle process coordinates.

[0048] The uniform periodic process coordinates are further segmented at intervals of 0.1, dividing the periodic range of 0 to 1 into 10 phase intervals, each corresponding to a periodic stage. Combining this with the aforementioned temporal mapping relationship, the laser diffraction signal is mapped to the corresponding uniform periodic process coordinates, and laser diffraction signals belonging to the same periodic stage are extracted, providing input data for subsequent comparison of signals within the same stage and selection of candidate equivalent signals.

[0049] The various events occurring within the time interval of the periodic phase are compared sequentially in the spray state sequence corresponding to the laser diffraction signal. If the order of occurrence of event types in two different spray cycles within the periodic phase is completely consistent and the number of occurrences of each event type is the same, then the corresponding spray state sequence is determined to have the same order of state changes, and the corresponding laser diffraction signal is determined as a candidate equivalent signal. If the order of occurrence of event types in two spray cycles is the same but the number of occurrences is different, then the laser diffraction signal with the same number of events as the reference spray cycle is selected as the candidate equivalent signal, where the reference spray cycle is the spray cycle whose total number of events within the periodic phase is closest to the median of the total number of events of all spray cycles.

[0050] By fusing drive status data, liquid supply status data, nozzle attitude status data, spray platform motion status data, and nozzle oscillation angle data, the start and end points of the spray cycle and the effective liquid supply time period can be uniformly identified. Using timestamp alignment, standardization processing, and dynamic time warping, spray cycles of different durations can be mapped to a unified cycle process coordinate system. By screening candidate equivalent signals based on cycle stages and the order of state changes, laser diffraction signals in different spray cycles can exhibit similar spray operation states, providing a consistent data foundation for subsequent comparison of overlapping particle size distributions, target signal screening, and spatial reconstruction.

[0051] Furthermore, the process of obtaining the overlapping particle size distribution vector includes, as follows: Figure 3 As shown: The background light reference signal is subtracted from the signals collected by the fixed diffraction path and the moving diffraction path in the candidate equivalent signals to obtain the net scattered light intensity vector of the fixed path and the net scattered light intensity vector of the moving path. The net scattered light intensity vector of the fixed path and the net scattered light intensity vector of the moving path are divided by their respective sampling volumes and then vertically concatenated to generate a joint light intensity vector. The submatrices corresponding to the fixed diffraction path and the moving diffraction path in the Mie scattering response matrix are vertically concatenated to generate a joint scattering response matrix. Under the constraint of the smoothness regularization factor, the non-negative least squares inversion algorithm is used to invert the joint light intensity vector using the joint scattering response matrix to output the overlapping particle size distribution vector.

[0052] When the ultrasonic sprayer stops spraying and the fixed and moving diffraction optical paths remain in measurement mode, light intensity data are continuously collected for 5 seconds each. At a sampling frequency of 100 Hz, 500 frames of light intensity data are collected for each diffraction optical path, and the average light intensity of each detection channel is calculated to obtain the background light reference signals for both the fixed and moving diffraction optical paths.

[0053] For each set of candidate equivalent signals at the overlapping measurement locations, the background light reference signal of the fixed diffraction path is subtracted from the light intensity vector collected by the fixed diffraction path to obtain the net scattered light intensity vector of the fixed path; the background light reference signal of the moving diffraction path is subtracted from the light intensity vector collected by the moving diffraction path to obtain the net scattered light intensity vector of the moving path. For channel light intensity values ​​that are less than 0 after background subtraction, the channel light intensity value is set to 0.

[0054] The sampling volume of the corresponding diffraction path is calculated based on the laser beam diameter and the effective measurement optical path. The calculation formula is as follows: in, Indicates the sampling volume. Indicates the diameter of the laser beam. The effective optical path length is indicated. In this embodiment, the laser beam diameter of both the fixed diffraction path and the moving diffraction path is 8 mm, the effective optical path length is 20 mm, and the corresponding sampling volume is 1005.3 cubic millimeters.

[0055] Before splicing the intensity vectors, cross-optical path metrological traceability and cross-calibration are performed: a standard polystyrene microparticle suspension with a known nominal particle size distribution is synchronously sprayed onto both the fixed and moving diffraction optical paths at the overlapping measurement positions. The linear operating range of the detectors in both optical paths is checked to avoid saturation, and channel-by-channel gain calibration, dark current background updates, and angular response calibration are performed to ensure that, under the same standard particle field, the relative deviation limit of the net scattered light intensity per unit volume output by the corresponding detection channels of the two optical paths is less than 2%. Under the premise of meeting the consistency limit, the net scattered light intensity vector of the fixed optical path is divided by the sampling volume of the fixed diffraction optical path to obtain the scattered light intensity vector per unit volume of the fixed optical path; the net scattered light intensity vector of the moving optical path is divided by the sampling volume of the moving diffraction optical path to obtain the scattered light intensity vector per unit volume of the moving optical path. The two scattered light intensity vectors per unit volume are then longitudinally spliced, with the fixed diffraction optical path first and the moving diffraction optical path second, to generate a joint intensity vector. Both the fixed diffraction optical path and the moving diffraction optical path include 32 detection channels, so the dimension of the joint intensity vector is 64×1.

[0056] Extract a fixed optical path scattering response submatrix with a dimension of 32×60 and a moving optical path scattering response submatrix with a dimension of 32×60 from the Mie scattering response matrix, and then vertically stitch them together with the fixed optical path scattering response submatrix on top and the moving optical path scattering response submatrix on the bottom to generate a joint scattering response matrix with a dimension of 64×60.

[0057] Using a non-negative least squares inversion algorithm with smoothness regularization constraints, solve the following equation: in, Represents the joint scattering response matrix. Represents the joint light intensity vector. This represents the particle size distribution vector to be solved. Represents a second-order difference matrix. This represents the smoothness regularization factor.

[0058] Set the candidate value of the smoothness regularization factor to 10. -6 For the 25 values ​​arranged logarithmically between 1 and 25 other values, calculate the residual norm and smoothness norm for each candidate value, and then use... The candidate values ​​corresponding to the point of maximum curvature of the curve are used as smoothness regularization factors.

[0059] exist Under the constraint that all elements are not less than 0, the solution is obtained iteratively. When the relative change between two consecutive iterations is less than 10... -6 Alternatively, the calculation can be stopped when the number of iterations reaches 1000, resulting in a 60-dimensional non-negative particle size distribution vector. Each element in this vector is divided by the sum of all elements until the sum equals 1. This normalized vector is then defined as the overlapping particle size distribution vector. Each element in the overlapping particle size distribution vector corresponds to a particle size interval, and the element value represents the volume fraction of droplets within that interval.

[0060] As another optional embodiment of the present invention, the non-negative least squares inversion algorithm can also be the basic standard non-negative least squares algorithm without smoothness regularization constraints. In this basic embodiment, the objective function is directly to minimize the scattering response fitting residual, i.e., to solve the following equation: exist The overlapping particle size distribution vector is obtained by iterative solution under the constraint that all elements are not less than 0. This basic implementation method is suitable for ideal measurement scenarios with extremely stable spray conditions, high signal-to-noise ratio of light intensity data, and minimal interference from ambient airflow. Its advantage lies in the fact that no regularization parameter is required. The optimization and tuning of the algorithm can further reduce its complexity and improve the speed of inversion calculation. However, when faced with unavoidable data noise in complex dynamic spraying operations, in order to effectively suppress abnormal fluctuation artifacts that may occur between adjacent particle size ranges, this invention preferably adopts the aforementioned non-negative least squares inversion algorithm with smoothness regularization constraints.

[0061] By subtracting the background light reference signals from the fixed and moving diffraction paths respectively, the influence of ambient light and detector baseline on light intensity data can be reduced. Normalizing the net scattered light intensity using the sampling volume ensures that the data collected by the two diffraction paths have uniform dimensions. Joint inversion of the light intensity vectors and scattering response sub-matrices of the two paths allows for the simultaneous use of scattering information from different detection channels. Non-negativity constraints ensure that the particle size distribution results meet the non-negativity condition of the volume fraction, while smoothness regularization and L-curve parameter selection limit abnormal fluctuations between adjacent particle size ranges, providing a unified data format for subsequent comparisons of particle size distributions across different spray cycles.

[0062] Furthermore, the process of selecting candidate equivalent signals with a difference value less than a set threshold as the target signal includes: For each candidate spray cycle containing a candidate equivalent signal, calculate the sum of the Euclidean distances between the overlapping particle size distribution vector corresponding to that candidate spray cycle and the overlapping particle size distribution vectors corresponding to all other candidate spray cycles containing the candidate equivalent signal. Use the sum of the Euclidean distances as the difference value corresponding to that candidate spray cycle. Summarize the difference values ​​corresponding to all candidate spray cycles, calculate the arithmetic mean of all difference values, and use the arithmetic mean as a set threshold. Extract candidate spray cycles with difference values ​​less than the set threshold as selected spray cycles, and determine the candidate equivalent signal corresponding to the selected spray cycle as the target signal.

[0063] For the same periodic phase, obtain the candidate equivalent signals. Each candidate spray cycle and the corresponding overlapping particle size distribution vector for each candidate spray cycle are given, where... The value is not less than 3, and each overlapping particle size distribution vector adopts the same particle size interval division method and has the same vector dimension.

[0064] The overlapping particle size distribution vector corresponding to the i-th candidate spray cycle is represented as... Calculate separately Overlapping particle size distribution vectors corresponding to the other candidate spray cycles The sum of all Euclidean distances is used as the difference value corresponding to the i-th candidate spray cycle. The calculation formula is: Summarize the differences for all candidate spray cycles, calculate the arithmetic mean of all differences, and directly determine this arithmetic mean as the set threshold T. Compare the differences for each candidate spray cycle one by one. With a set threshold T. When the difference value When the difference is less than the set threshold T, the corresponding candidate spray cycle is determined as the selected spray cycle; when the difference value is less than the set threshold T, the candidate spray cycle is determined as the selected spray cycle. If the value is greater than or equal to a set threshold T, the corresponding candidate spray cycle is not selected as the chosen spray cycle. Based on the cycle number of the selected spray cycle, the laser diffraction signal belonging to each selected spray cycle is extracted from the candidate equivalent signal corresponding to the same cycle stage, and the extracted laser diffraction signal is determined as the target signal.

[0065] Compare the differences for each candidate spray cycle one by one. With set threshold When the difference value Less than the set threshold When the difference value is found, the corresponding candidate spray cycle is determined as the selected spray cycle; when the difference value is found, the corresponding candidate spray cycle is determined as the selected spray cycle. Greater than or equal to the set threshold When this happens, the corresponding candidate spray cycle is not selected as the chosen spray cycle.

[0066] Based on the selected spray cycle number, laser diffraction signals belonging to each selected spray cycle are extracted from the candidate equivalent signals corresponding to the same cycle stage, and the extracted laser diffraction signals are determined as target signals. The target signals retain the corresponding candidate spray cycle number, cycle stage number, target measurement position number, and acquisition time, which are used to subsequently combine target signals from different target measurement positions within the same cycle stage into spatial reconstruction data.

[0067] By calculating the sum of the Euclidean distances between the overlapping particle size distribution vectors corresponding to each candidate spray cycle and the remaining candidate spray cycles, the deviation of each candidate spray cycle from the overall particle size distribution in the same stage can be quantified. A set threshold is determined using the arithmetic mean of all differences, allowing for adaptive selection criteria based on the current candidate data. Selecting candidate spray cycles with differences below the set threshold reduces the number of spray cycles with large particle size distribution deviations entering the spatial recombination process, ensuring better consistency between the spray state and particle size distribution corresponding to the target signal, and providing stable data input for subsequent spatial droplet particle size distribution inversion.

[0068] Furthermore, the process of combining the data into spatially reconstructed data includes: Acquire nozzle attitude state data and spray operation platform motion state data when collecting the laser diffraction signal; calculate the three-dimensional spatial coordinates and spray attitude angle of each target measurement position based on the nozzle attitude state data and spray operation platform motion state data; map the target signal corresponding to each target measurement position to a three-dimensional spatial grid using the three-dimensional spatial coordinates and the spray attitude angle to obtain node signal data; perform grid interpolation and spatial matrix splicing on each node signal data to generate the spatial reconstruction data.

[0069] Based on the acquisition time of the target signal, the nozzle pitch angle and yaw angle at the corresponding time are extracted from the nozzle attitude state data, and the platform's three-dimensional position coordinates at the corresponding time are extracted from the spray operation platform's motion state data. The nozzle pitch angle, yaw angle, and platform's three-dimensional position coordinates are all recorded using the same time reference as the laser diffraction signal. When the acquisition time of the target signal coincides with the sampling time of the state data, the corresponding state parameters are directly extracted; when the acquisition time of the target signal is between two adjacent state data sampling times, linear interpolation is performed based on the time ratio between the acquisition time and the two adjacent sampling times to obtain the state parameters corresponding to the target signal acquisition time.

[0070] The spatial coordinates of each target measurement position in the pre-built 3D coordinate system are used as local position coordinates. A pitch rotation matrix is ​​constructed based on the nozzle pitch angle, and a yaw rotation matrix is ​​constructed based on the nozzle yaw angle. The two rotation matrices are then multiplied in the order of pitch rotation first, followed by yaw rotation, to obtain the spatial rotation matrix. The 3D position coordinates of the spraying platform are used as translation vectors. The local position coordinates of the target measurement position are then rotated using the spatial rotation matrix, and the translation vectors are superimposed to obtain the 3D spatial position coordinates of the target measurement position corresponding to the target signal acquisition.

[0071] The nozzle pitch angle and nozzle yaw angle are combined to form the spray attitude angle. The nozzle pitch angle represents the angle between the spray direction and the horizontal plane, and the nozzle yaw angle represents the angle between the projection of the spray direction onto the horizontal plane and the positive X-axis. A correspondence is established between the spray attitude angle and the target signal, the acquisition time, and the three-dimensional spatial coordinates to characterize the spatial spray direction of the nozzle during target signal acquisition.

[0072] A three-dimensional spatial grid is established within the spray area, with the grid spacing set to 20 mm in the X, Y, and Z axes. Based on the minimum and maximum coordinate values ​​of the spray area in the three coordinate axes, the number of grid nodes in each coordinate axis is determined, and corresponding X-axis, Y-axis, and Z-axis indices are assigned to each grid node.

[0073] For each target measurement location, the coordinate distance between it and adjacent grid nodes is calculated based on its three-dimensional spatial coordinates. When the three-dimensional spatial coordinates of the target measurement location coincide with the coordinates of a certain grid node, the target signal is directly written into that grid node, and the target signal, injection attitude angle, acquisition time, cycle stage number, and target measurement location number are associated and stored to obtain the node signal data corresponding to that grid node.

[0074] When the three-dimensional spatial coordinates of the target measurement location do not coincide with the grid node coordinates, the grid coordinates on both sides of the three-dimensional spatial coordinates are determined in the X-axis, Y-axis and Z-axis directions respectively. The grid coordinates in the three coordinate axes are combined to form 8 adjacent grid nodes surrounding the target measurement location.

[0075] Based on the relative position of the target measurement location within adjacent grid intervals along the X, Y, and Z axes, linear interpolation weights are calculated for each of the three coordinate axes. These linear interpolation weights are then multiplied to obtain the trilinear interpolation weights of the target measurement location relative to each adjacent grid node. Each trilinear interpolation weight corresponding to an adjacent grid node is not less than 0, and the sum of the eight trilinear interpolation weights is 1.

[0076] Considering that the laser diffraction signal originates from a finite cylindrical measurement volume with a diameter of 8 mm and an optical path of 20 mm, its essence is the line-of-sight integral along the beam path, that is, the scattered light intensity of each detection channel satisfies the spatial integral model: in Let k be the spatial sensitivity kernel of the k-th channel. For spatial point distribution, This represents the error term. Without performing 3D tomography inversion, when the target signal is multiplied by its corresponding trilinear interpolation weights and mapped to eight adjacent grid nodes, the resulting data is not the actual physical point value at the node, but rather composite data centered on the current grid node and encompassing a 20mm integral effect along the optical path. The target signal is a scattered light intensity vector arranged in the order of the detection channels. Therefore, during trilinear interpolation, the scattered light intensity values ​​corresponding to each detection channel are weighted and calculated to obtain the node signal data containing the optical path convolution effect for each adjacent grid node.

[0077] When the same grid node receives node signal data mapped from multiple target measurement positions, the scattered light intensity values ​​of the same detection channel are weighted and summed according to the trilinear interpolation weights corresponding to each target measurement position. The weighted summation result is then divided by the sum of the weights received by the grid node to obtain the scattered light intensity data corresponding to that grid node.

[0078] Following the order of increasing X-axis coordinates, increasing Y-axis coordinates, and increasing Z-axis coordinates, spatial matrix stitching is performed on the node signal data within the same period. To avoid mathematical inconsistencies arising from the superposition of light intensity interpolation and nonlinear inversion, no spatial interpolation is performed on the original scattered light intensity. Only nodes containing actual measured light intensity data are saved as valid spatial reconstruction data; grid nodes that did not acquire the target signal are marked as null values ​​and will be spatially reconstructed in the particle size distribution feature dimension after inversion. For grid nodes with fewer than 8 surrounding valid grid nodes, no interpolation is performed on the grid node, and it is marked as an invalid grid node.

[0079] The target signal, interpolated signal data, 3D spatial coordinates, jet attitude angle, acquisition time, and period stage number corresponding to all effective grid nodes are associated and stored to generate spatial reconstruction data. This spatial reconstruction data is established separately for each period stage and arranged according to the index order of the 3D spatial grid. Each effective grid node corresponds to a set of 3D spatial coordinates, a set of jet attitude angles, and a set of scattered light intensity data arranged in the order of the detection channels, which are used for subsequent particle size inversion of the scattered light intensity data of each grid node.

[0080] By combining nozzle attitude data with spray platform motion data, the local coordinates of each target measurement location can be converted into three-dimensional spatial coordinates at the time of target signal acquisition, establishing a correspondence between target signal, spray attitude angle, and spatial position. A three-dimensional spatial grid is used to uniformly map the scattered light intensity data from different target measurement locations, and trilinear interpolation is used to supplement missing nodes within the effective measurement area. This generates spatially reconstructed data arranged according to spatial indexes and periodic stages, providing structurally consistent data input for subsequent particle size inversion of each grid node and generation of spatial droplet size distribution.

[0081] Furthermore, the process of generating the spatial droplet size distribution of the ultrasonic sprayer at the corresponding periodic stage includes: Scattered light intensity data of each target measurement location is extracted from the spatial reconstructed data. A non-negative least squares inversion algorithm is used to invert the scattered light intensity data of each target measurement location to obtain the particle size distribution vector of each target measurement location. The spatial coordinates of each target measurement location are obtained, and each particle size distribution vector is mapped to the corresponding spatial coordinates and spliced ​​according to the spatial position relationship to generate a global spatial particle size distribution matrix. The time interval is determined according to the start and end points and the corresponding periodic stage. The particle size data located in the time interval in the global spatial particle size distribution matrix is ​​extracted to generate the spatial droplet particle size distribution of the ultrasonic sprayer in the corresponding periodic stage.

[0082] From the spatially reconstructed data, according to the periodic stage number and the target measurement location number, the scattered light intensity data, acquisition time, spatial coordinates, and diffraction path identifier corresponding to each target measurement location are extracted. Based on the diffraction path identifier, the scattering response submatrix corresponding to the scattered light intensity data is extracted from the Mie scattering response matrix.

[0083] The net scattered light intensity vector is obtained by subtracting the background light reference signal of the corresponding diffraction path from the scattered light intensity data corresponding to each target measurement position. For channel light intensity values ​​that are less than 0 after background subtraction, the corresponding channel light intensity value is set to 0. The net scattered light intensity vector is divided by the sampling volume of the corresponding diffraction path to obtain the net scattered light intensity vector per unit volume, and the net scattered light intensity vector per unit volume is used as the input data for the non-negative least squares inversion algorithm.

[0084] For each target measurement location, under the constraint that all elements in the particle size distribution vector are not less than 0, the aforementioned non-negative least squares inversion algorithm (preferably a non-negative least squares inversion algorithm with smoothness regularization constraint, or the basic standard non-negative least squares algorithm under ideal low noise conditions) is used to invert the net scattered light intensity vector per unit volume using the corresponding scattering response submatrix, and the particle size distribution vector corresponding to the target measurement location is obtained.

[0085] The particle size distribution vector comprises 60 elements, each corresponding to one of 60 equally logarithmically spaced particle size intervals within the range of 1 micrometer to 500 micrometers. After the inversion calculation is completed, each element in the particle size distribution vector is divided by the sum of all elements to obtain the normalized particle size distribution vector. Each element in the normalized particle size distribution vector represents the volume fraction of droplets within the corresponding particle size interval at the target measurement location.

[0086] Based on the target measurement location number, the three-dimensional spatial coordinates corresponding to each target measurement location are extracted from the spatial reconstruction data. The normalized particle size distribution vector of each target measurement location is mapped to the corresponding three-dimensional spatial coordinates to form spatial particle size node data containing three-dimensional spatial coordinates, particle size range, and volume fraction.

[0087] Spatial particle size node data within the same period are arranged in ascending order of X-axis coordinate, Y-axis coordinate, and Z-axis coordinate. The number of grid nodes along the X-axis, Y-axis, and Z-axis, and the number of particle size intervals are denoted as follows: , , And 60, constructing dimensions as The global spatial particle size distribution matrix.

[0088] In the global spatial particle size distribution matrix, element F(a,b,c,r) represents the droplet volume fraction corresponding to the r-th particle size interval at the a-th X-axis grid node, the b-th Y-axis grid node, and the c-th Z-axis grid node. For grid nodes with target measurement locations, their independently inverted normalized particle size distribution vectors and their uncertainties / covariances are directly written into the grid node. For grid nodes in effective measurement areas without actual measurements, adjacent effective spatial particle size nodes surrounding the grid node are selected, and suitable characteristic parameters (such as characteristic particle size) of the particle size distribution of these actual nodes are extracted. , , The feature parameters (or the volume percentage of each particle size range) are spatially estimated using a trilinear interpolation method to generate the spatial particle size distribution data corresponding to the grid node.

[0089] For the q-th spray cycle, extract the start time of the spray cycle from the start and end points. and termination time Let the starting value of the corresponding periodic stage in the coordinate system of the unified periodic process be... The termination value is ,in, Then the time interval corresponding to this periodic stage is: Based on the acquisition time corresponding to each spatial particle size node, the global spatial particle size distribution matrix within the time interval at which the acquisition time is located is extracted. For the same spatial grid node containing multiple particle size distribution vectors within the time interval, the effective selected spray cycle sample size N within that spatial grid node is first calculated. If the sample size N is less than the preset minimum number of effective cycles (e.g., 3 times), the node is determined to lack statistical significance and is marked as an invalid mask. If the sample size N is greater than or equal to the minimum number of effective cycles, the mean (or median) of the volume fraction in each particle size interval is calculated and used as the central particle size distribution vector of the spatial grid node in the corresponding cycle stage. Simultaneously, the periodic variance of each cycle in the corresponding particle size interval is calculated to construct a statistical confidence interval based on the sample size N.

[0090] The center particle size distribution vectors and statistical characteristics corresponding to each spatial grid node within the time interval are arranged according to their three-dimensional spatial positional relationships to generate a set of spatial droplet particle size distribution features with statistical confidence for the ultrasonic sprayer in the corresponding period stage. This feature set includes not only the three-dimensional spatial coordinates of each spatial grid node, the particle size range of each particle size interval, and the center particle size distribution vector, but also the complete output of the corresponding effective period set identifier, sample size N, the phase center and phase width of the current period stage, the period variance (or confidence interval), and the invalid mask matrix for nodes with missing measurements or insufficient sample size. When the proportion of effective grid nodes across the entire domain falls below a preset threshold, an alarm is triggered and the output of the distribution results for the current period stage is stopped.

[0091] It should be noted that, limited by the effective measurement optical path integration effect of 20mm, the actual spatial resolution of this composite distribution in the beam propagation direction is about 20mm. When facing a strong gradient non-uniform spray field, its grid node values ​​will include cross-contamination and local bias across multiple grids.

[0092] By performing non-negative least squares inversion on the scattered light intensity in the spatially reconstructed data, the particle size distribution vector corresponding to each target measurement location can be obtained, ensuring that the inversion result satisfies the non-negativity constraint of particle size volume fraction. Mapping the particle size distribution vector to the corresponding three-dimensional spatial coordinates and performing spatial stitching yields a global spatial particle size distribution matrix containing information on spatial location and particle size intervals. Combining the start and end points of the spray cycle and the time intervals corresponding to the cycle stages, particle size data within the same cycle stage can be extracted, generating a spatial droplet particle size distribution with a correspondence between time stages and spatial locations.

[0093] This embodiment establishes a correspondence between target measurement positions, laser diffraction signals, nozzle oscillation angles, and spray cycle stages by unifying the three-dimensional coordinate system, synchronizing the time reference, and using multi-source state data. Joint inversion is performed at overlapping measurement positions using fixed and moving diffraction optical paths, and target signals are filtered based on differences in overlapping particle size distribution. This effectively identifies and eliminates abnormal spray cycles caused by significant fluctuations in the global output state of spray sources such as the liquid supply system or transducers, ensuring that the target signals entering spatial reconstruction have a consistent reference state in the time dimension, thereby reducing the interference of time baseline drift on the spatial distribution reconstruction results. Subsequently, combining nozzle attitude and platform motion state, the scattered light intensity data of each target measurement position is mapped to a three-dimensional spatial grid. Through particle size inversion, grid interpolation, and spatial matrix stitching, the spatial droplet particle size distribution under the corresponding cycle stage is generated, realizing the correlation characterization of droplet particle size in time stage and spatial position.

[0094] Example 2 A method for measuring droplet size in an ultrasonic sprayer based on laser diffraction, before combining target signals corresponding to the same period and located at each target measurement position into spatially reconstructed data, further includes a time-varying optical path transmission calibration process for the target signals based on constraints of drug liquid characteristic parameters, specifically including: The moving diffraction optical path is controlled to reach the overlapping measurement position before and after acquiring the laser diffraction signal at the target measurement position, and synchronously acquires the calibration diffraction signal of the same period stage with the fixed diffraction optical path. Simultaneously, the characteristic parameters of the sprayed drug solution are acquired, including the complex refractive index, viscosity, surface tension, and concentration. To address the differences in optical response caused by changes in drug composition, the complex refractive index of the drug solution (including the real part and the imaginary absorption part) is extracted, substituted into the Mie scattering theory model, and the response coefficients of each detection channel for different particle size ranges are recalculated, thereby dynamically updating the Mie scattering response matrices of the fixed and moving diffraction optical paths.

[0095] To address the differences in physical distribution caused by changes in the hydrodynamic properties of the drug solution, based on the viscosity of the drug solution... Surface tension ,density And the Onezog number of the current drug solution is calculated based on the baseline particle size D (taken as 100 micrometers). The Onezoglu number of pure water under the same conditions. For reference, combined with the volume concentration parameters of the drug solution Construct sampling volume flux correction coefficient The specific calculation formula is as follows: in, This is a dimensionless empirical index for nozzle flux. This is the empirical extinction coefficient. Parameter and Before the equipment leaves the factory, droplet flux calibration experiments can be conducted using a phase Doppler interferometer on pure water and standard suspensions of different concentrations. The results can be obtained through polynomial fitting (for conventional ultrasonic atomizing nozzles). The value typically ranges from 0.2 to 0.5. This coefficient is used... Compensate for the deviation in particle number density within the effective measurement optical path caused by changes in spray cone angle and fluid viscosity.

[0096] To address the baseline drift of the instrument during measurement, for each detection channel, the pre-transfer coefficient is calculated using the calibration diffraction signals acquired before and after the target measurement position. With post-transfer coefficient The pre-transfer coefficient is the ratio of the net light intensity value of the detection channel corresponding to the fixed diffraction optical path to the net light intensity value of the moving diffraction optical path in the pre-calibration signal; the post-transfer coefficient is calculated in the same way.

[0097] Based on the target signal acquisition time By performing linear time interpolation on the pre-transfer coefficient and the post-transfer coefficient, the transfer coefficient at the target time is obtained: The target signal is subjected to time drift correction of hardware channel response by combining the target time transmission coefficient, and the corrected net light intensity is multiplied by the sampling volume flux correction coefficient. The final corrected target signal is obtained. Subsequently, the corrected target signal is combined with the updated Mie scattering response matrix for inversion and spatial reconstruction.

[0098] A method for measuring droplet size in an ultrasonic sprayer based on laser diffraction, wherein the process of mapping the laser diffraction signal within each spraying cycle to the corresponding cycle stage using the acquisition time includes: Acquire droplet velocity data corresponding to each target measurement position; determine the spray direction at the corresponding acquisition time based on the ultrasonic sprayer nozzle attitude state data, and calculate the propagation path length along the spray direction from the nozzle to the target measurement position; calculate the propagation time required for pesticide droplets to propagate from the nozzle outlet to the target measurement position based on the propagation path length and droplet velocity data; subtract the propagation time from the laser diffraction signal acquisition time to obtain the spray source time when the corresponding pesticide droplets actually leave the nozzle; map the laser diffraction signal to the corresponding spray cycle stage based on the spray source time and the start and end points of the spray cycle.

[0099] In Example 1, the laser diffraction signal is bound to the corresponding node of the spray state sequence according to the acquisition time of the laser diffraction signal, and the periodic stage of the laser diffraction signal is determined according to the unified periodic process coordinates.

[0100] Based on Example 1, this embodiment addresses the issue that the droplet state at the detection location lags behind the actual spraying state of the nozzle due to the spatial distance between the nozzle and the crop target during pesticide spraying. It compensates for the propagation delay of the laser diffraction signal at the corresponding periodic stage to improve the accuracy of droplet state matching between different spraying cycles.

[0101] The remaining steps follow the processing method described in Example 1.

[0102] Specifically, while acquiring laser diffraction signals at each target measurement location via a moving diffraction optical path, pesticide droplet velocity data at the corresponding locations are simultaneously acquired. The droplet velocity data characterizes the time it takes for pesticide droplets to travel from the nozzle exit to the crop canopy or the target measurement area.

[0103] Based on the laser diffraction signal acquisition time, the corresponding jet direction is obtained from the nozzle attitude state data, and combined with the nozzle spatial position and the target measurement position spatial position, the actual propagation path length of the droplets along the jet direction is calculated.

[0104] The propagation time of the droplets is calculated based on the propagation path length and droplet velocity data. Droplet velocity data corresponding to each target measurement location is obtained. Since the time-series velocity obtained from a single fixed location is an Eulerian observation and cannot be directly equated to the spatial motion history of a single droplet, this embodiment constructs a Lagrange trajectory model based on particle size grouping and aerodynamics to calculate the actual propagation delay.

[0105] The spray direction at the corresponding acquisition time is determined based on the ultrasonic sprayer nozzle attitude state data, and the actual propagation path length L from the nozzle to the target measurement position along the spray direction is calculated.

[0106] The initial jet velocity of the droplets is obtained by a reference velocity measuring node pre-set at the nozzle exit. Simultaneously, a laser Doppler velocimeter was used to obtain the arrival velocity of particle size groups at the target measurement location. For characteristic particle size of The median fog droplet, based on the Lagrange differential equation of motion of spherical particles decelerated by drag in the air: Combined with boundary conditions and Construct the length along the injection path Spatial velocity decay field .

[0107] The time required for pesticide droplets to travel from the nozzle outlet to the target measurement location By the spatial velocity attenuation field The result is obtained by spatial line integration along the path: The definite integral is solved using Gauss-Legendal numerical integration or other numerical integration methods to obtain a high-precision droplet propagation time. .

[0108] The time of laser diffraction signal acquisition is denoted as . ,according to: Obtain the spray source time when the corresponding pesticide droplets actually leave the nozzle. .

[0109] Based on the relationship between the jet source time and the start and end points of each spray cycle, the laser diffraction signal is reclassified to the corresponding spray cycle, and the corresponding cycle stage is determined according to the unified cycle process coordinates.

[0110] The periodic phase, which has been corrected for propagation delay, is used to replace the periodic phase determined based on the acquisition time. Laser diffraction signals in the same spray state phase during different spray cycles are extracted as candidate equivalent signals for subsequent droplet size distribution calculation.

[0111] A method for measuring droplet size in an ultrasonic sprayer based on laser diffraction includes, in addition to acquiring the spatial coordinates of the target measurement location, collecting laser diffraction signals, and generating the spatial droplet size distribution, an adaptive supplementary measurement step based on the spatial droplet size distribution characteristics. Specifically, this includes: Target measurement positions are set according to the initial grid spacing, and the initial spatial droplet size distribution is obtained by inversion based on the laser diffraction signals collected at each target measurement position. For different spatial grid units, supplementary measurement indicators for the corresponding spatial region are calculated based on the particle size inversion error, repeatability stability, and the degree of change in particle size distribution between adjacent positions. Spatial grid units with supplementary measurement indicators exceeding a set threshold are extracted as regions to be refined. Supplementary target measurement positions are added to the regions to be refined, and the moving diffraction optical path is controlled to collect supplementary laser diffraction signals. Particle size inversion and spatial matrix stitching are performed again using the supplementary measurement results to update the spatial droplet size distribution. The above refinement process is repeated until the spatial grid units meet the measurement accuracy requirements or reach the preset minimum grid spacing.

[0112] In Example 1, multiple target measurement positions are set according to a preset spatial spacing. The laser diffraction signals collected at each target measurement position are used to perform particle size inversion, and the particle size distribution vector corresponding to each position is mapped to a three-dimensional spatial grid to generate a spatial droplet particle size distribution.

[0113] Based on Example 1, this embodiment addresses the issue of differences in droplet size distribution in different spatial regions during pesticide spraying due to factors such as nozzle structure, spray posture, airflow disturbance, and crop canopy obstruction. It adaptively supplements the measurement area based on the initial spatial droplet size distribution results.

[0114] The remaining steps follow the processing method described in Example 1.

[0115] During the initial measurement, a three-dimensional spatial grid was set according to the spray area range, and multiple initial target measurement positions were determined according to the initial grid spacing. The moving diffraction optical path was controlled to sequentially reach each initial target measurement position, and the laser diffraction signal at the corresponding position was collected. Based on the candidate equivalent signal extraction, target signal screening, and particle size inversion process in Example 1, the particle size distribution vector corresponding to each target measurement position was obtained.

[0116] The particle size distribution vectors corresponding to the measurement locations of each target are mapped to a three-dimensional spatial grid, and the initial spatial droplet particle size distribution is generated by splicing the spatial matrices.

[0117] For each spatial grid cell, the particle size inversion residual, the degree of change in particle size distribution during repeated measurements, and the difference in particle size distribution between adjacent spatial locations are obtained within that region.

[0118] Among them, the particle size inversion residual is used to characterize the degree of matching between the laser diffraction signal inversion result and the actual collected scattered light intensity; the particle size distribution dispersion of repeated measurements is used to characterize the stability of the measurement results in this region; and the difference in particle size distribution between adjacent locations is used to characterize the trend of droplet size variation in this region.

[0119] The three parameters are subjected to Min-Max normalization to map each parameter to the [0,1] interval.

[0120] The normalized particle size inversion residual is denoted as The dispersion of particle size distribution in repeated measurements is denoted as... The difference in particle size distribution between adjacent locations is denoted as .

[0121] Supplementary measurement indicators Calculated according to the equal-weighted fusion method: The set threshold is set to 0.3, that is, when When the value is greater than 0.3, the spatial grid cell is determined as the region to be refined.

[0122] When the supplementary measurement index of a certain spatial grid cell exceeds the set threshold, the spatial grid cell is identified as a region to be refined, and supplementary target measurement positions are added at the center and boundary of the region.

[0123] The moving diffraction optical path is controlled to reach the measurement positions of each supplementary target, and the supplementary laser diffraction signal is collected. Then, the particle size distribution vector corresponding to the measurement position of the supplementary target is obtained according to the signal screening and particle size inversion process in Example 1.

[0124] The particle size distribution vector obtained from the supplementary measurement replaces the original spatial interpolation result, and is then spliced ​​with the particle size distribution vector corresponding to the existing target measurement location in a spatial matrix to obtain the updated spatial droplet particle size distribution.

[0125] After completing one round of supplementary measurements, the supplementary measurement parameters for each spatial grid cell are recalculated. For areas where there are still significant grain size variations or measurement uncertainties, the local grid spacing is further reduced, and additional supplementary target measurement locations are added for measurement.

[0126] When all grid cells within the spatial region meet the set accuracy requirements, the supplementary measurement is stopped, and the final spatial droplet size distribution is taken as the measurement result.

[0127] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction, characterized in that, include: The spatial coordinates of multiple target measurement positions and overlapping measurement positions within the spray area are obtained. The laser diffraction signals of the target measurement positions and overlapping measurement positions and the corresponding acquisition time are collected. The nozzle swing angle data of the ultrasonic sprayer in multiple spray cycles are also obtained. The start and end points of multiple spray cycles are determined based on the nozzle swing angle data. The laser diffraction signals in each spray cycle are mapped to the corresponding cycle stage using the acquisition time. The laser diffraction signals in the same cycle stage in different spray cycles are extracted as candidate equivalent signals. A non-negative least squares inversion algorithm is used to invert the candidate equivalent signals corresponding to the overlapping measurement locations to obtain the overlapping particle size distribution vector. The difference between the overlapping particle size distribution vectors corresponding to different spray cycles is calculated, and the candidate equivalent signals with a difference value less than a set threshold are selected as the target signals. The target signals corresponding to the same cycle stage and located at each target measurement location are combined into spatial reconstructed data. The non-negative least squares inversion algorithm is used to invert the spatially reconstructed data. The inversion results are stitched together using the spatial coordinates of the target measurement location to generate the spatial droplet size distribution of the ultrasonic sprayer in the corresponding period stage.

2. The method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction according to claim 1, characterized in that, The process of acquiring the spatial coordinates of multiple target measurement positions within the spray area, as well as overlapping measurement positions, collecting laser diffraction signals and corresponding acquisition times of the target measurement positions and overlapping measurement positions, and acquiring nozzle swing angle data of the ultrasonic sprayer in multiple spray cycles includes: extracting the coordinate values ​​of the target measurement positions as the spatial coordinates in a pre-built three-dimensional coordinate system; deploying a fixed diffraction optical path at the overlapping measurement positions and controlling a moving diffraction optical path to sequentially reach each of the target measurement positions; synchronously acquiring light energy distribution data as the laser diffraction signal through the fixed diffraction optical path and the moving diffraction optical path, and extracting the timestamp corresponding to the light energy distribution data as the acquisition time; calculating the Mie scattering response matrix based on the optical parameters of the fixed diffraction optical path and the moving diffraction optical path; and continuously acquiring the real-time angle sequence of the ultrasonic sprayer's rotating shaft using an angle sensor, and using the real-time angle sequence as the nozzle swing angle data.

3. The method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction according to claim 1, characterized in that, The process of extracting laser diffraction signals from the same period stage in different spray cycles as candidate equivalent signals includes: acquiring the driving state data, liquid supply state data, nozzle attitude state data, and spray operation platform motion state data of the ultrasonic sprayer; inputting the driving state data, liquid supply state data, and nozzle swing angle data into a state event recognition model; and extracting the timestamps of spray start-up events, liquid supply state transition events, and nozzle swing reversal events as start and end points; inputting the driving state data, liquid supply state data, nozzle attitude state data, spray operation platform motion state data, and nozzle swing angle data into a multi-source data fusion model; and extracting the liquid supply flow rate using the liquid supply state data, with the liquid supply flow rate being greater than a preset value. The flow rate threshold is used as a data validity constraint. Data that meets the data validity constraint is sampled, aligned, and dimensionless, and a multi-dimensional state feature matrix corresponding to the spray cycle is output as the spray state sequence. Using the acquisition time, each laser diffraction signal is bound to the corresponding node of the spray state sequence to generate time-series mapping signal data. The spray state sequences of different spray cycles are input into the time-series alignment model, mapped to the unified periodic process coordinates based on the start and end points, and divided at fixed intervals to obtain multiple phase intervals as periodic stages. Using the time-series mapping signal data, laser diffraction signals that are in the same periodic stage and whose corresponding spray state sequences have the same state change order are extracted as candidate equivalent signals.

4. The method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction according to claim 1, characterized in that, The process of obtaining the overlapping particle size distribution vector includes: subtracting the background light reference signal from the signals collected by the fixed diffraction path and the moving diffraction path in the candidate equivalent signals to obtain the net scattered light intensity vector of the fixed path and the net scattered light intensity vector of the moving path; dividing the net scattered light intensity vector of the fixed path and the net scattered light intensity vector of the moving path by the corresponding sampling volume and then concatenating them vertically to generate a joint light intensity vector; concatenating the submatrices corresponding to the fixed diffraction path and the moving diffraction path in the Mie scattering response matrix vertically to generate a joint scattering response matrix; and, under the constraint of the smoothness regularization factor, using the non-negative least squares inversion algorithm to invert the joint light intensity vector using the joint scattering response matrix to output the overlapping particle size distribution vector.

5. The method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction according to claim 1, characterized in that, The process of selecting candidate equivalent signals with a difference value less than a set threshold as target signals includes: for each candidate spray cycle containing candidate equivalent signals, calculating the sum of the Euclidean distances between the overlapping particle size distribution vector corresponding to the candidate spray cycle and the overlapping particle size distribution vectors corresponding to all other candidate spray cycles containing candidate equivalent signals, and using the sum of Euclidean distances as the difference value corresponding to the candidate spray cycle; summing the difference values ​​corresponding to all candidate spray cycles, calculating the arithmetic mean of all difference values, and determining the arithmetic mean as the set threshold; extracting candidate spray cycles with a difference value less than the set threshold as selected spray cycles, and determining the candidate equivalent signals corresponding to the selected spray cycles as target signals.

6. The method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction according to claim 1, characterized in that, The process of combining the data into spatial reconstructed data includes: acquiring nozzle attitude state data and spray operation platform motion state data when the laser diffraction signal is collected; calculating the three-dimensional spatial position coordinates and spray attitude angles of each target measurement position based on the nozzle attitude state data and spray operation platform motion state data; mapping the target signals corresponding to each target measurement position to a three-dimensional spatial grid using the three-dimensional spatial position coordinates and the spray attitude angles to obtain node signal data; and performing grid interpolation and spatial matrix splicing on each node signal data to generate the spatial reconstructed data.

7. The method for measuring the droplet size of an ultrasonic sprayer based on laser diffraction according to claim 1, characterized in that, The process of generating the spatial droplet size distribution of the ultrasonic sprayer in the corresponding period stage includes: extracting the scattered light intensity data of each target measurement position from the spatial reconstruction data; performing inversion calculation on the scattered light intensity data of each target measurement position using a non-negative least squares inversion algorithm to obtain the particle size distribution vector of each target measurement position; obtaining the spatial coordinates of each target measurement position; mapping each particle size distribution vector to the corresponding spatial coordinates; and splicing them according to the spatial position relationship to generate a global spatial particle size distribution matrix; determining the time interval based on the start and end points and the corresponding period stage; extracting the particle size data located in the time interval from the global spatial particle size distribution matrix; and generating the spatial droplet size distribution of the ultrasonic sprayer in the corresponding period stage.

Images