A multi-source data acquisition device and method for low-altitude logistics unmanned aerial vehicles

Abstract

The application provides a multi-source data acquisition device and method for low-altitude logistics unmanned aerial vehicles, relates to the technical field of data acquisition, and is characterized in that different independent sampling periods are allocated to multiple types of sensors, each sensor is driven to operate asynchronously at an adaptive frequency, each sampling data is marked by time stamp analysis, a sliding window boundary is used as a unified screening reference, a sampling ready time stamp of each sample is compared with a window starting time and an ending time, effective samples of each high-speed dynamic signal channel are subjected to time axis recalculation operation according to respective average acquisition delay amounts through an effective sample set, current pulsation feature time points and rotating speed pulse feature time points in a window are extracted from time-aligned multi-source data sets, the time domain phase relationship of the current pulsation feature time points and the rotating speed pulse feature time points is compared with a theoretical fixed phase value of a corresponding motor, and quality analysis marking is performed, so that the application can solve the time axis alignment problem of multi-source heterogeneous data in a sliding time window.

Description

Technical Field

[0001] This invention proposes a multi-source data acquisition device and method for low-altitude logistics drones, which relates to the field of data acquisition technology, specifically to the field of multi-source data acquisition technology for low-altitude logistics drones. Background Technology

[0002] In the application scenario of health prediction and graded safety control during low-altitude logistics drone flight, airborne multi-source data acquisition is a fundamental prerequisite for estimating short-term failure probability. Logistics drones typically carry various heterogeneous sensors such as current, vibration, rotational speed, and temperature sensors. These sensors are connected to the flight control computing platform through different physical interfaces and communication protocols, with sampling frequencies ranging from tens of hertz to thousands of hertz.

[0003] When the health prediction module uses a sliding time window to extract feature vectors, the aforementioned multi-source heterogeneous sensors suffer from factors such as sampling frequency mismatch, differences in signal transmission paths due to distributed physical layout, and inconsistent analog-to-digital conversion times among channels. This results in relative time offsets in the current, vibration, rotational speed, and temperature data collected within the same window. This time axis misalignment causes the components of the feature vector input to the health prediction model to not correspond to the same physical moment, thus significantly reducing the reliability of the failure probability estimation.

[0004] In existing data acquisition schemes, the global unified synchronous triggering method cannot eliminate the channel deviation due to the different transmission delays of the synchronization signal over long distances, while the post-interpolation resampling method causes the early fault characteristic frequencies in the high-frequency vibration signal to be smoothly lost during the interpolation process. Summary of the Invention

[0005] This invention provides a multi-source data acquisition device and method for low-altitude logistics drones to solve the above-mentioned problems: This invention proposes a multi-source data acquisition device and method for low-altitude logistics drones, the method comprising: S1. By allocating different independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, each sensor is driven to run asynchronously at its own adapted frequency, and each sampled data is timestamped and marked to obtain the original sampling sequence. S2. Using the original sampling sequence and its dual timestamp information, and taking the sliding window boundary as a unified screening benchmark, the sampling ready timestamp of each sample is compared and analyzed with the window start and end times to obtain the effective sample set and its completeness statistics. S3. By using the effective sample set, perform time axis reduction operation on the effective samples of each high-speed dynamic signal channel according to their respective average acquisition delay to obtain a time-aligned multi-source dataset. S4. Extract the current pulsation feature time and speed pulse feature time within the window by time-aligned multi-source dataset, compare and verify the time-domain phase relationship between the two with the theoretical fixed phase value of the corresponding motor model, and perform quality analysis and marking to obtain multi-source data acquisition marking information.

[0006] Furthermore, the system includes: The original module is used to allocate different independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, drive each sensor to run asynchronously at its own adapted frequency, and perform timestamp analysis and marking on each sampled data to obtain the original sampling sequence; The time analysis module is used to compare and analyze the sampling ready timestamp of each sample with the window start and end times using the original sampling sequence and its carried dual timestamp information, with the sliding window boundary as a unified screening criterion, to obtain the effective sample set and its completeness statistics. The time alignment module is used to perform time axis reduction on the effective samples of each high-speed dynamic signal channel according to their respective average acquisition delay, thereby obtaining a time-aligned multi-source dataset. The acquisition and marking module is used to extract the current pulsation feature time and speed pulse feature time within the window by aligning the multi-source dataset with time, compare and verify the time-domain phase relationship between the two with the theoretical fixed phase value of the corresponding motor model, and perform quality analysis and marking to obtain multi-source data acquisition and marking information.

[0007] The beneficial effects of this invention are as follows: This invention solves the time axis misalignment problem caused by differences in sampling frequency, distributed physical layout, and different signal transmission paths of multi-source heterogeneous sensors in low-altitude logistics drones under airborne conditions, ensuring that each component of the feature vector input to the health prediction module strictly corresponds to the same physical moment; the acquisition delay characteristics of each channel are quantified through a dual timestamp marking mechanism, providing a physical basis for subsequent time alignment; outlier samples caused by abnormal acquisition delays are eliminated through sliding window filtering, reducing the interference of erroneous data on the health prediction model; self-verification of multi-source data time consistency is achieved through physical phase cross-validation, improving the closed-loop guarantee capability of the data acquisition system for time synchronization accuracy; and quality grading marking enables the health prediction module to adopt differentiated processing strategies under different data quality conditions, taking into account both the continuity and accuracy of data reliability prediction. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of a multi-source data acquisition method for low-altitude logistics drones. Detailed Implementation

[0009] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0010] In one embodiment of the present invention, a multi-source data acquisition device and method for low-altitude logistics drones are provided, the method comprising: S1. By allocating different independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, each sensor is driven to run asynchronously at its own adapted frequency, and each sampled data is timestamped and marked to obtain the original sampling sequence. S2. Using the original sampling sequence and its dual timestamp information, and taking the sliding window boundary as a unified screening benchmark, the sampling ready timestamp of each sample is compared and analyzed with the window start and end times to obtain the effective sample set and its completeness statistics. S3. By using the effective sample set, perform time axis reduction operation on the effective samples of each high-speed dynamic signal channel according to their respective average acquisition delay to obtain a time-aligned multi-source dataset. S4. Extract the current pulsation feature time and speed pulse feature time within the window using time-aligned multi-source datasets. Compare and verify the time-domain phase relationship between the two with the theoretical fixed phase value of the corresponding motor model, and perform quality analysis and labeling to obtain multi-source data acquisition labeling information, such as... Figure 1 As shown.

[0011] Multi-source data acquisition refers to the technical process of simultaneously acquiring operational status data of key components of a low-altitude logistics drone from multiple types of sensors, such as current, vibration, rotation speed, and temperature, during the drone's flight, providing multi-dimensional raw input information for health prediction models.

[0012] The channel configuration table is a data structure stored in the flight control platform's memory. It records configuration information such as the physical interface type, sampling frequency parameters, channel number, and signal type classification of each sensor channel. This information is used by the acquisition coprocessor to perform differentiated scheduling and management of each channel.

[0013] High-speed dynamic signal channel group refers to a collection of sensor channels operating at a sampling rate of several kHz, including signals such as motor phase current, bus current, and triaxial vibration acceleration, used to capture high-frequency transient change characteristics to identify early fault signs such as bearing wear, blade damage, and abnormal current ripple.

[0014] The medium-speed status signal channel group refers to a collection of sensor channels operating at sampling rates from hundreds of Hz to 1 kHz, including signals such as propeller speed and aircraft attitude angular rate, used to track real-time changes in the aircraft's motion status.

[0015] The low-speed trend signal channel group refers to a set of sensor channels operating at a sampling rate of tens of Hz, including signals such as battery voltage, motor temperature, battery temperature, and driver temperature, used to monitor the changing trend of physical quantities that drift slowly over time.

[0016] Independent sampling period refers to the different sampling interval time parameters configured by the acquisition coprocessor for each channel according to the change rate requirements of different signal categories, so that high-speed channels are sampled at high frequency and low-speed channels are sampled at low frequency, avoiding resource waste or loss of high-frequency information caused by uniform sampling rate.

[0017] Asynchronous operation means that each sensor channel triggers its acquisition action autonomously according to its own independently configured sampling period. The channels do not depend on the same clock beat to start synchronously and work in parallel with different phases and frequencies on the time axis.

[0018] The trigger timestamp refers to the current value of the global time base counter latched by the input capture unit when the acquisition coprocessor sends a synchronous sampling trigger signal. It is used to mark the physical trigger time of this acquisition action.

[0019] The sampling ready timestamp refers to the current value of the global time base counter, which is latched again by the input capture unit when the analog-to-digital converter completes the conversion of the current sample and sets the conversion complete flag. This value is used to mark the ready time when the sample data is actually available.

[0020] The dual timestamp label is a time stamp information composed of a trigger timestamp and a sampling ready timestamp. It is combined with the channel number and sampling sequence number of the corresponding sampling data and appended to the original sampling data to quantify the acquisition delay characteristics of the sample from triggering to data readiness.

[0021] A time-stamped data frame is a structured data unit that packages the raw sampled data, double timestamp labels, channel number, and sample sequence number according to a predetermined data format. It is used to transmit data samples carrying complete time-series information between the acquisition coprocessor and the health prediction module.

[0022] A sliding window is a data analysis interval divided into fixed time lengths on continuous time series data. The window slides forward gradually over time, and the health prediction module extracts feature vectors based on the effective samples within the current window.

[0023] The window time boundary parameter is a parameter pair consisting of the start and end times of the current sliding window, which serves as the time interval benchmark for uniformly filtering all channel data samples.

[0024] Valid samples in the initial screening refer to data samples whose sampling ready timestamps fall strictly within the interval between the start and end times of the current window, and are retained for subsequent timeline calculation operations.

[0025] Out-of-bounds abnormal samples refer to data samples whose sampling ready timestamp is less than the start time of the window or greater than the end time of the window, and are therefore removed because they do not belong to the current window in terms of time.

[0026] The completeness statistic is the percentage of valid samples in each channel within the current window relative to the theoretical number of samples for that channel. It is used to quantify the sufficiency of data for each channel within the current window.

[0027] The acquisition delay is the difference between the sampling ready timestamp and the trigger timestamp of the same valid sample. It represents the time required for the channel to go from triggering sampling to data conversion readiness. It is determined by the signal conditioning circuit delay, analog-to-digital conversion time, and bus transmission delay.

[0028] The average acquisition delay is calculated in channels and is a statistical value obtained by arithmetically averaging the acquisition delays of all valid samples within the current window. This value is used as a uniform compensation amount for the time axis reduction operation of that channel within the current window.

[0029] The time axis reduction operation is the process of shifting the time coordinate of each valid sample forward from its sampling ready time to its corresponding synchronization trigger time. This is achieved by subtracting the average acquisition delay of the channel from the sampling ready timestamp, so that the data of each channel within the same window are unified to a consistent physical time reference on the time coordinate.

[0030] After alignment, the time axis is reduced from the sampling ready time to the synchronization trigger time of the sample data. The sampled values ​​remain unchanged, but the time axis is reset.

[0031] A time-aligned multi-source dataset is a collection of data obtained by merging and summarizing aligned samples from all channels within the same window, where the data from each channel points to a consistent physical time sequence on the time coordinate.

[0032] The time-domain position corresponding to the peak value of the phase current pulsation in the current channel within the current pulsation characteristic time window is formed by the periodic current fluctuations generated during the motor commutation process.

[0033] The time domain position corresponding to the rising edge of the speed pulse signal in the speed channel within the speed pulse characteristic time window is marked by the pulse signal generated by the propeller for each revolution.

[0034] The measured time-domain phase relationship value, which is the relative phase offset between the current pulsation characteristic time sequence and the speed pulse characteristic time sequence within the same time window, reflects the degree of measured time alignment between the current and speed signals.

[0035] The theoretical fixed phase value is the fixed phase relationship between the current pulsation frequency and the speed, which is determined by the number of pole pairs of the motor. It serves as the reference value for cross-verification and does not change with load and speed.

[0036] The deviation tolerance threshold is pre-stored in the parameter storage area of ​​the flight control platform as a judgment threshold, which is used to measure whether the deviation between the measured phase relationship value and the theoretical fixed phase value is within an acceptable range.

[0037] The phase deviation is the difference between the measured time-domain phase relationship value and the theoretical fixed phase value. Its magnitude reflects the data time alignment accuracy, and its positive or negative sign indicates the leading or lagging direction of the current channel relative to the speed channel.

[0038] The time consistency verification result is determined by comparing the phase deviation with the deviation tolerance threshold. The result indicates whether the alignment accuracy of the multi-source data in the current window on the time axis meets the input requirements of the health prediction model.

[0039] The correction direction indicator determines the lead or lag relationship between the current channel and the speed channel based on the positive or negative sign of the phase deviation, and is used to guide the adjustment direction of the acquisition delay compensation parameters of this channel in subsequent windows.

[0040] The multi-source data acquisition labeling information includes a comprehensive labeling output that includes the current window data quality level, time consistency deviation indicator value, and correction direction information. This is used to control how the data in this window is output to the health prediction module.

[0041] The working principle and technical effects of the above technical solution are as follows: This invention allocates differentiated independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, enabling each sensor to operate asynchronously and collect raw data according to its own adapted frequency; by marking each sampled data with a dual time tag of trigger timestamp and sampling ready timestamp, each data sample carries complete timing information; by uniformly filtering the sampling ready timestamps of all samples through the sliding window boundary, samples whose timestamps fall within the window interval are retained as valid samples, and abnormal samples that exceed the boundary are eliminated; by calculating the average acquisition delay of valid samples by channel and performing time axis reduction operation, the data of each channel within the same window are unified to a consistent physical time reference on the time coordinate; finally, by extracting the current pulsation characteristic time and speed pulse characteristic time within the window, the measured phase relationship between the two is compared and verified with the theoretical fixed phase value of the motor, and the data quality is graded and marked by combining the completeness statistics of each channel.

[0042] This system addresses the time axis misalignment issue caused by differences in sampling frequency, distributed physical layout, and signal transmission paths of multi-source heterogeneous sensors in low-altitude logistics drones under airborne conditions, ensuring that each component of the feature vector input to the health prediction module strictly corresponds to the same physical moment. A dual-timestamping mechanism quantifies the acquisition delay characteristics of each channel, providing a physical basis for subsequent time alignment. A sliding window filtering method eliminates outliers caused by abnormal acquisition delays, reducing the interference of erroneous data on the health prediction model. Physical phase cross-validation enables self-verification of multi-source data time consistency, improving the closed-loop guarantee capability of the data acquisition system for time synchronization accuracy. Quality grading and labeling allow the health prediction module to employ differentiated processing strategies under different data quality conditions, balancing the continuity and accuracy of data reliability prediction.

[0043] In one embodiment of the present invention, S1 includes: By assigning differentiated independent sampling period parameters to the high-speed dynamic signal channel group, the medium-speed status signal channel group, and the low-speed trend signal channel group through the channel configuration table of the flight control platform, the sampling period configuration results of each channel group are obtained. By configuring the sampling period, each sensor is driven to run asynchronously at its own adapted frequency, so that the high-speed channel group runs at the highest sampling rate, the medium-speed channel group runs at the medium sampling rate, and the low-speed channel group runs at the lowest sampling rate, thus obtaining the original acquisition data of each channel under its own sampling cycle. By performing timestamp analysis and marking on each raw data acquisition, a trigger timestamp is marked when the corresponding raw sampling data is triggered for acquisition, and a sampling ready timestamp is marked when the corresponding raw sampling data is ready for analog-to-digital conversion, thus obtaining the raw sampling sequence of each channel carrying dual timestamp labels.

[0044] The working principle and technical effect of the above technical solution are as follows: This invention assigns differentiated independent sampling period parameters to the high-speed dynamic signal channel group, the medium-speed status signal channel group, and the low-speed trend signal channel group through the channel configuration table of the flight control platform; the sampling period configuration results drive each sensor to operate asynchronously at its own adapted frequency, so that the high-speed channel group operates at the highest sampling rate to capture the transient change characteristics of vibration and current, the medium-speed channel group operates at a medium sampling rate to track the real-time changes of rotational speed and attitude, and the low-speed channel group operates at the lowest sampling rate to monitor the slow drift trend of temperature and voltage; by marking the trigger timestamp and the sampling ready timestamp at the trigger time and the conversion ready time of each raw acquisition data respectively, each data sample carries a dual time sequence label.

[0045] This approach resolves the contradiction between the inability of a single sampling rate to simultaneously capture high-frequency signals and consume computational resources in low-frequency channels, enabling each sensor to operate at its optimal rate. This improves the overall efficiency of multi-source data acquisition and reduces redundant invalid data. By using dual timestamp marking, the time offset introduced by differences in physical layout, signal conditioning delay, and bus transmission time in each channel is transformed into quantifiable acquisition delay characteristic values, providing basic data for subsequent time alignment compensation.

[0046] In one embodiment of the present invention, the step of performing timestamp analysis and marking on each raw data acquisition, marking a trigger timestamp when the corresponding raw sampling data is triggered for acquisition, and marking a sampling ready timestamp when the corresponding raw sampling data is ready for analog-to-digital conversion, to obtain the raw sampling sequence of each channel carrying dual timestamp tags, includes: The sampling trigger time of each sensor channel is monitored in real time through the channel configuration table of the flight control platform. When the acquisition coprocessor sends a synchronous sampling trigger signal, the input capture unit of the coprocessor latches the current value of the global time base counter to obtain the trigger timestamp. By monitoring the completion time of analog-to-digital conversion for each sensor channel in real time, when the analog-to-digital converter completes the conversion of the current sample and sets the conversion completion flag, the input capture unit of the coprocessor latches the current value of the global time base counter again to obtain the sampling ready timestamp. By combining the trigger timestamp and the sampling ready timestamp according to the channel number and sampling sequence number of the corresponding sampling data, the two are combined into a double timestamp label. The double timestamp label and the original sampling data are then packaged together into a time-stamped data frame to obtain the original sampling sequence of each channel carrying the double timestamp label.

[0047] Specifically, the sampling trigger time of each sensor channel is monitored in real time through the channel configuration table of the flight control platform. When the acquisition coprocessor sends a synchronous sampling trigger signal, the input capture unit of the coprocessor latches the current value of the global time base counter to obtain the trigger timestamp, including: The flight control platform monitors the timing of the synchronous trigger signal of each sensor channel in real time through the channel configuration table. When the acquisition coprocessor sends a synchronous sampling trigger pulse to any high-speed dynamic signal channel according to the sampling period configuration result, the interrupt controller inside the coprocessor detects the rising edge signal of the trigger pulse to obtain the detection signal of the trigger event. By triggering the event detection signal, the input capture unit inside the coprocessor performs a latching operation within the first timer clock cycle after the interrupt response, latching the current value of the global time base counter to obtain the original value of the trigger timestamp corresponding to the trigger event; By triggering the original timestamp value, recording the channel number and sampling sequence number corresponding to the triggering event, and associating and binding the two with the original timestamp value, the inherent relationship between the sample and the triggering time can be obtained.

[0048] The working principle and technical effect of the above technical solution are as follows: The present invention monitors the timing of the synchronous trigger signal of each sensor channel in real time through the channel configuration table of the flight control platform. When the acquisition coprocessor sends a synchronous sampling trigger pulse, the interrupt controller detects the rising edge signal of the trigger pulse to obtain the detection signal of the trigger event. The detection signal triggers the input acquisition unit to perform a latching operation in the first timer clock cycle after the interrupt response to latch the current value of the global time base counter and obtain the original value of the trigger timestamp. The channel number and sampling sequence number corresponding to the trigger event are recorded and associated with the original value of the trigger timestamp.

[0049] This invention solves the problem of traditional software-based timestamping methods where the timestamp recording time lags behind the physical trigger time due to interrupt response delays and task scheduling jitter. By directly latching the time base counter value at the rising edge of the trigger pulse through the hardware input capture unit, the timestamp recording accuracy is improved to the level of the timer clock cycle, eliminating the impact of operating system scheduling uncertainty on the accuracy of trigger time recording. By binding the channel number and sampling sequence number to the trigger timestamp, a unique mapping relationship between the trigger time and the corresponding data sample is ensured.

[0050] In one embodiment of the present invention, S2 includes: By using the original sampling sequence and its carried dual timestamp information, the start and end times of the current sliding window are read as the time boundaries for unified filtering, and the window time boundary parameters are obtained. By comparing and analyzing the sampling ready timestamp of each sample with the start and end times of the window using the window time boundary parameter, samples whose sampling ready timestamps fall strictly within the window interval are retained, and abnormal samples whose timestamps exceed the window boundaries are removed, thus obtaining the initial screening effective samples for each channel within the current window. By classifying and summarizing the initial screening valid samples by channel, the percentage of valid samples in the current window for each channel relative to the theoretical sampling quantity of that channel is calculated, thus obtaining the completeness statistics for each channel and the set of valid samples for each channel.

[0051] The working principle and technical effect of the above technical solution are as follows: This invention reads the start and end times of the current sliding window as a unified screening time boundary by using the original sampling sequence and its carried dual timestamp information; by comparing the sampling ready timestamp of each sample with the start and end times of the window through this time boundary, samples whose sampling ready timestamps strictly fall within the window interval are retained as initially screened valid samples, and out-of-bounds abnormal samples whose timestamps exceed the window boundary are removed; by classifying and summarizing the initially screened valid samples by channel, the percentage of the number of valid samples in the current window for each channel relative to the theoretical sampling number of that channel is calculated.

[0052] This solution addresses the issue where the actual time of data from some channels within the same sliding window deviates from the physical range of the window due to different acquisition delays across channels. By using the sampling readiness timestamp as a filtering criterion, abnormal samples that do not belong to the current window in terms of time are eliminated, ensuring that each data sample participating in health prediction belongs to the current analysis window in terms of time. Through the calculation of the integrity statistics of each channel, the data acquisition system can identify the sufficiency of data for each channel within the current window.

[0053] In one embodiment of the present invention, the step of comparing and analyzing the sampling ready timestamp of each sample with the start and end times of the window using window time boundary parameters, retaining samples whose sampling ready timestamps strictly fall within the window interval, and removing abnormal samples whose timestamps exceed the window boundaries, to obtain the initial screening effective samples for each channel within the current window, includes: By using the window start time and window end time in the window time boundary parameters, the sampling ready timestamps of all samples in the current window are read point by point to obtain the timestamp value of each sample. The sampling is judged by comparing the timestamp value of each sample with the window start time. If the sampling ready timestamp is greater than or equal to the window start time, the next step of judgment is carried out. If the sampling ready timestamp is less than the window start time, the sample is judged as an out-of-bounds abnormal sample on the left and is removed, thus obtaining the sample set after the left boundary is filtered. After filtering the sample set by the left boundary, the sampling ready timestamp of each sample in the set is compared with the window end time. If the sampling ready timestamp is less than or equal to the window end time, the sample is determined to be a valid sample within the window and retained in the initial screening valid sample set. If the sampling ready timestamp is greater than the window end time, the sample is determined to be an out-of-bounds abnormal sample on the right and is removed. This process obtains the initial screening valid samples for each channel within the current window.

[0054] Specifically, by using the window start time and window end time in the window time boundary parameters, the sampling ready timestamps of all samples within the current window are read point by point to obtain the timestamp value of each sample, including: By using the window start time and window end time in the window time boundary parameters, the time interval range covered by the current window is determined, and the data range boundary to be traversed and scanned is obtained. By traversing and scanning all the original sampling sequences of all channels within the time interval covered by the current window through the data range boundary, the time-stamped data frames of each sample are accessed one by one to obtain the access entry of each sample data frame. By parsing each sample data frame, the binary value of the sampling ready timestamp field is read from the fixed offset address in the data frame, and the binary value is converted into a microsecond-level time unit to obtain the timestamp value of each sample. By parsing the timestamp value of each sample and reading the corresponding channel number and sampling sequence number from the data frame, the timestamp value is associated with and stored with the channel number and sampling sequence number to obtain the timestamp value and channel affiliation information of each sample in the current window.

[0055] The working principle and technical effect of the above technical solution are as follows: The present invention determines the time interval range covered by the current window by using the window start time and end time in the window time boundary parameters to obtain the data range boundary to be traversed and scanned; the entire original sampling sequence of all channels within the window time interval is traversed and scanned by using the data range boundary and the time-stamped data frame of each sample is accessed one by one; the binary value of the sampling ready timestamp field is read from the fixed offset address by parsing the data frame of each sample and converted into microsecond-level time units; the channel number and sampling sequence number corresponding to the sample are read and associated with the timestamp value for storage.

[0056] The problem of difficult timestamp field location caused by differences in data frame formats of different channels was solved. The timestamp information in heterogeneous data frames was uniformly extracted by using a fixed offset address parsing method. By uniformly converting the sampling ready timestamp into microsecond-level time units, the problem of inconsistent units caused by the difference in timing accuracy within each sensor was eliminated, making timestamps from different channels comparable. By storing timestamp values ​​in association with channel numbers and sampling sequence numbers, a dual index relationship of sample time position and channel affiliation within the window was established.

[0057] In one embodiment of the present invention, S3 includes: By extracting the trigger timestamp and sampling ready timestamp carried by each valid sample from the set of valid samples in each channel and calculating the difference between the two, the single acquisition delay of each sample is obtained. By calculating the average acquisition delay of all valid samples in the current window based on the single acquisition delay of each sample, the average acquisition delay of each channel in the current window is obtained. By performing time axis reduction on the effective samples of each high-speed dynamic signal channel through the average acquisition delay of each channel, the time coordinate of each sample is reduced from its sampling ready time to its corresponding synchronous trigger time, and the aligned samples with the time coordinates of each channel are unified to the trigger time reference are obtained. By merging and summarizing the aligned samples by channel, signals within the same window point to the same physical time sequence on the time coordinate, thus obtaining a time-aligned multi-source dataset.

[0058] The working principle and technical effect of the above technical solution are as follows: This invention extracts the trigger timestamp and sampling ready timestamp carried by each valid sample from the valid sample set of each channel and calculates the difference between the two to obtain the single acquisition delay of each sample; it calculates the average acquisition delay of all valid samples in the current window by calculating the average acquisition delay of each channel by calculating the single acquisition delay of each sample by channel; it performs time axis reduction operation on the valid samples of each high-speed dynamic signal channel by using the average acquisition delay, so that the time coordinate of each sample is reduced from its sampling ready time to its corresponding synchronization trigger time; by merging and summarizing the aligned samples after reduction by channel, the data of each channel in the same window points to the same physical time sequence on the time coordinate.

[0059] This invention addresses the issue of inconsistent data readiness times across channels at the same trigger moment, caused by differences in analog-to-digital conversion times of various sensors and varying delays in signal conditioning circuits. By retroactively calculating the time coordinate of each sample from its actual readiness time to the synchronous trigger time, the data from all channels within the same window are aligned with the same physical trigger reference on the time axis, eliminating relative time offsets between different channels. Furthermore, by calculating the average acquisition delay per channel, the time axis compensation value is adapted to the individual delay characteristics of each channel, avoiding over-compensation or under-compensation for channels with significant delay differences using a uniform compensation amount.

[0060] In one embodiment of the present invention, the step of performing time axis reduction operation on the effective samples of each high-speed dynamic signal channel by means of the average acquisition delay of each channel, and reducing the time coordinate of each sample backward from its sampling ready time to its corresponding synchronization trigger time, to obtain aligned samples in each channel with time coordinates unified to the trigger time reference, includes: By using the average acquisition delay of each channel within the current window, the sampling ready timestamp of each valid sample in the high-speed dynamic signal channel is read to obtain the original time coordinate value of each sample. By subtracting the average acquisition delay of the channel from the original time coordinate value, the reduced time coordinate value is obtained, so that the time coordinate of the sample is shifted forward from its actual data ready time to its corresponding synchronization trigger time. By replacing the original time coordinate of the sample with the time coordinate value after reduction and keeping the sampling value unchanged, the time axis reduction operation of a single valid sample is completed, and a single aligned sample is obtained. By performing a reduction operation on all valid samples of the high-speed dynamic signal channel one by one, the valid samples of all channels within the same window are aligned to the physical time reference corresponding to the start time of the window on the reduced time coordinates, thus obtaining the aligned samples of each channel with the time coordinates unified to the trigger time reference.

[0061] Specifically, by using the average acquisition delay of each channel within the current window, the sampling readiness timestamp of each valid sample in the high-speed dynamic signal channel is read to obtain the original time coordinate value of each sample, including: By using the average acquisition delay of each channel within the current window, the effective sample set of each high-speed dynamic signal channel is traversed and accessed one by one, and the time-stamped data frame of each effective sample is obtained, thus obtaining the data frame access entry of each effective sample. By parsing each valid sample data frame, the sampling ready timestamp field is extracted from the data frame to obtain the original time coordinate value of each valid sample. By parsing each valid sample data frame, the channel number field is extracted from the data frame. Based on the channel number, the high-speed dynamic signal channel category to which the sample belongs is determined. The average acquisition delay is read from the storage area corresponding to the channel to obtain the pairing relationship between the original time coordinate value of each sample and the average acquisition delay of its corresponding channel.

[0062] The working principle and technical effect of the above technical solution are as follows: This invention traverses and accesses the effective sample set of each high-speed dynamic signal channel by the average acquisition delay of each channel within the current window and obtains the time-stamped data frame of each effective sample one by one; by parsing each effective sample data frame, the sampling ready timestamp field is extracted from the data frame to obtain the original time coordinate value of each effective sample; by parsing each effective sample data frame, the channel number field is extracted, the high-speed dynamic signal channel category to which the sample belongs is confirmed according to the channel number, and the average acquisition delay is read from the storage area corresponding to the channel to obtain the pairing relationship between the original time coordinate value of each sample and the average acquisition delay of its corresponding channel.

[0063] This solution addresses the issue of increased computational overhead caused by the dispersed storage locations of average acquisition delay values ​​across channels, necessitating multiple retrievals during the recalculation process. By using the channel number field as an index, time coordinate extraction and corresponding delay value reading are performed simultaneously before recalculation, reducing the number of data accesses and bus occupancy time. Furthermore, by establishing a pairing relationship between the original time coordinate values ​​and the average acquisition delay values ​​of the corresponding channels, the accuracy of the recalculation operation is ensured, and the exacerbation of time axis misalignment due to delay value mismatches is avoided.

[0064] In one embodiment of the present invention, S4 includes: By extracting the feature time corresponding to the peak value of phase current pulsation in the current channel within the window using time-aligned multi-source datasets, and simultaneously extracting the feature time corresponding to the rising edge of the speed pulse in the speed channel, the current pulsation feature time sequence and the speed pulse feature time sequence are obtained. By using the characteristic time sequence of current pulsation and the characteristic time sequence of rotational speed pulse, the relative phase offset between the two in the time domain of the window is calculated, and the measured time domain phase relationship value of the current window is obtained. The measured time-domain phase relationship value is compared and verified with the theoretical fixed phase value of the corresponding motor pole pair number, and the deviation between the two is calculated to obtain the time consistency deviation indication value. By combining the time consistency deviation indicator value with the integrity statistics of each channel, the data quality of the current window is comprehensively judged and graded to obtain multi-source data acquisition labeling information. This labeling information is used to control the output method of the current window data to the health prediction module.

[0065] The working principle and technical effect of the above technical solution are as follows: This invention extracts two characteristic time sequences by extracting the characteristic time corresponding to the peak value of phase current pulsation in the current channel and the characteristic time corresponding to the rising edge of speed pulse in the speed channel from the time-aligned multi-source dataset. The relative phase offset in the time domain of the window is calculated using these two sequences to obtain the measured time domain phase relationship value of the current window. The measured phase relationship value is compared and verified with the theoretical fixed phase value of the number of pole pairs of the corresponding motor model, and the deviation is calculated to obtain the time consistency deviation indication value. The deviation indication value is combined with the integrity statistics of each channel to comprehensively judge and classify the data quality of the current window to obtain multi-source data acquisition marking information.

[0066] This solves the problem that relying solely on timestamp information cannot verify whether data from different channels are truly physically synchronized after time axis reduction. By utilizing the fixed physical coupling relationship between current pulsation and rotational speed as an independent verification method, it provides an airborne self-verification capability for time alignment accuracy that does not depend on external reference signals. By comprehensively judging the phase deviation indication value and the channel integrity statistics value, the data quality classification considers both the data sufficiency of each channel and the time consistency level between multiple channels, avoiding misclassification caused by a single dimension judgment.

[0067] In one embodiment of the present invention, the step of comparing and verifying the measured time-domain phase relationship value with the theoretical fixed phase value of the corresponding model motor pole pair number, calculating the deviation between the two, and obtaining a time consistency deviation indication value includes: The theoretical fixed phase value corresponding to the current number of motor pole pairs is read from the motor parameter configuration table stored on the flight control platform by the measured time-domain phase relationship value of the current window, and the phase verification reference value is obtained. The phase deviation of the current window is obtained by calculating the difference between the measured time-domain phase relationship value and the theoretical reference value. The preset deviation tolerance threshold value is read from the parameter storage area of ​​the flight control platform by the phase deviation amount and the two are compared and judged. If the phase deviation amount is less than or equal to the deviation tolerance threshold value, the time consistency verification of the data in the window is determined to be passed. If the phase deviation amount is greater than the deviation tolerance threshold value, the time consistency verification of the data in the window is determined to be failed, and the time consistency verification result is obtained. By analyzing the time consistency verification failure result and the specific numerical direction of the phase deviation, the time lead or lag relationship between the current channel and the speed channel is identified based on the positive or negative sign of the phase deviation. The lead or lag relationship is used as the correction direction indicator for the subsequent window acquisition delay compensation parameters, thereby obtaining the time consistency deviation indication value and the corresponding correction direction information.

[0068] Specifically, the theoretical fixed phase value corresponding to the current number of motor pole pairs is read from the motor parameter configuration table stored on the flight control platform using the measured time-domain phase relationship value of the current window, to obtain the phase verification reference value, including: By accessing the motor model identifier field stored in the non-volatile storage area of ​​the flight control platform through the measured time-domain phase relationship value of the current window, the model code of the motor currently carried by the UAV is read, and the current motor model identifier is obtained. Using the current motor model identifier, retrieve the record entry that matches the identifier from the motor parameter configuration table pre-stored in the non-volatile memory area to obtain the complete parameter record of the motor model. By parsing the motor parameter records, the number of motor pole pairs can be extracted from the records to obtain the current number of motor pole pairs. By substituting the number of pole pairs of the motor into the theoretical relationship between current pulsation frequency and speed, the theoretical fixed phase value of the motor at the current speed is obtained as a phase verification benchmark, thus obtaining a phase verification benchmark value that matches the current motor.

[0069] The working principle and technical effect of the above technical solution are as follows: This invention accesses the motor model identifier field stored in the non-volatile storage area of ​​the flight control platform through the measured time-domain phase relationship value of the current window to read the model code of the motor currently carried by the UAV and obtain the current motor model identifier; it retrieves the matching record entry from the motor parameter configuration table pre-stored in the non-volatile storage area through the identifier to obtain the complete parameter record of the motor model; it extracts the motor pole pair number parameter from the motor parameter record by parsing the record to obtain the current motor pole pair number value; and it calculates the theoretical fixed phase value of the motor at the current speed by substituting the pole pair number parameter into the theoretical relationship between the current pulsation frequency and the rotation speed, as a phase verification benchmark.

[0070] This invention solves the problem of the lack of universality of theoretical fixed phase values ​​due to the different pole pair numbers of different motor models. By reading the current motor model from the non-volatile storage area and retrieving the corresponding pole pair number parameter, the calibration benchmark value is automatically matched with the actual motor model without manual configuration or online calculation. By substituting the pole pair number parameter into the theoretical formula to obtain the theoretical phase value, the calibration benchmark can be adaptively updated with changes in speed, avoiding the judgment deviation caused by fixed value calibration when the speed changes. By using this theoretical phase value as the calibration benchmark, a unified quantitative evaluation of time consistency under different motor models and different speed conditions is achieved.

[0071] In one embodiment of the present invention, the system includes: The original module is used to allocate different independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, drive each sensor to run asynchronously at its own adapted frequency, and perform timestamp analysis and marking on each sampled data to obtain the original sampling sequence; The time analysis module is used to compare and analyze the sampling ready timestamp of each sample with the window start and end times using the original sampling sequence and its carried dual timestamp information, with the sliding window boundary as a unified screening criterion, to obtain the effective sample set and its completeness statistics. The time alignment module is used to perform time axis reduction on the effective samples of each high-speed dynamic signal channel according to their respective average acquisition delay, thereby obtaining a time-aligned multi-source dataset. The acquisition and marking module is used to extract the current pulsation feature time and speed pulse feature time within the window by aligning the multi-source dataset with time, compare and verify the time-domain phase relationship between the two with the theoretical fixed phase value of the corresponding motor model, and perform quality analysis and marking to obtain multi-source data acquisition and marking information.

[0072] The working principle and technical effects of the above technical solution are as follows: This system allocates differentiated independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, enabling each sensor to operate asynchronously and collect raw data according to its own adapted frequency; by marking each sampled data with a dual time tag of trigger timestamp and sampling ready timestamp, each data sample carries complete timing information; by uniformly filtering the sampling ready timestamps of all samples through the sliding window boundary, samples whose timestamps fall within the window interval are retained as valid samples, and abnormal samples that exceed the boundary are removed; by calculating the average acquisition delay of valid samples by channel and performing time axis reduction operation, the data of each channel within the same window are unified to a consistent physical time reference on the time coordinate; finally, by extracting the current pulsation characteristic time and speed pulse characteristic time within the window, the measured phase relationship between the two is compared and verified with the theoretical fixed phase value of the motor, and the data quality is graded and marked by combining the completeness statistics of each channel.

[0073] This system addresses the time axis misalignment issue caused by differences in sampling frequency, distributed physical layout, and signal transmission paths of multi-source heterogeneous sensors in low-altitude logistics drones under airborne conditions. It ensures that each component of the feature vector input to the health prediction module strictly corresponds to the same physical moment. A dual-timestamping mechanism quantifies the acquisition delay characteristics of each channel, providing a physical basis for subsequent time alignment. A sliding window filtering method eliminates outliers caused by abnormal acquisition delays, reducing the interference of erroneous data on the health prediction model. Physical phase cross-validation enables self-verification of multi-source data time consistency, improving the closed-loop guarantee capability of the data acquisition system for time synchronization accuracy. Quality grading and labeling allow the health prediction module to employ differentiated processing strategies under different data quality conditions, balancing the continuity and accuracy of data reliability prediction.

[0074] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A multi-source data acquisition method for low-altitude logistics drones, characterized in that, The method includes: S1. By allocating different independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, each sensor is driven to run asynchronously at its own adapted frequency, and each sampled data is timestamped and marked to obtain the original sampling sequence. S2. Using the original sampling sequence and its dual timestamp information, and taking the sliding window boundary as a unified screening benchmark, the sampling ready timestamp of each sample is compared and analyzed with the window start and end times to obtain the effective sample set and its completeness statistics. S3. By using the effective sample set, perform time axis reduction operation on the effective samples of each high-speed dynamic signal channel according to their respective average acquisition delay to obtain a time-aligned multi-source dataset. S4. Extract the current pulsation feature time and speed pulse feature time within the window by time-aligned multi-source dataset, compare and verify the time-domain phase relationship between the two with the theoretical fixed phase value of the corresponding motor model, and perform quality analysis and marking to obtain multi-source data acquisition marking information.

2. The multi-source data acquisition method for low-altitude logistics drones according to claim 1, characterized in that, S1 includes: By assigning differentiated independent sampling period parameters to the high-speed dynamic signal channel group, the medium-speed status signal channel group, and the low-speed trend signal channel group through the channel configuration table of the flight control platform, the sampling period configuration results of each channel group are obtained. By configuring the sampling period, each sensor is driven to run asynchronously at its own adapted frequency, so that the high-speed channel group runs at the highest sampling rate, the medium-speed channel group runs at the medium sampling rate, and the low-speed channel group runs at the lowest sampling rate, thus obtaining the original acquisition data of each channel under its own sampling cycle. By performing timestamp analysis and marking on each raw data collection, a trigger timestamp is marked when the corresponding raw data collection is triggered, and a sampling ready timestamp is marked when the corresponding raw data analog-to-digital conversion is ready, thus obtaining the raw sampling sequence.

3. The multi-source data acquisition method for low-altitude logistics drones according to claim 2, characterized in that, The process involves timestamping each raw data point, marking a trigger timestamp when the corresponding raw data is triggered for acquisition, and marking a sampling readiness timestamp when the corresponding raw data is ready for analog-to-digital conversion, to obtain the raw sampling sequence, including: The sampling trigger time of each sensor channel is monitored in real time through the channel configuration table of the flight control platform. When the acquisition coprocessor sends a synchronous sampling trigger signal, the input capture unit of the coprocessor latches the current value of the global time base counter to obtain the trigger timestamp. By monitoring the completion time of analog-to-digital conversion for each sensor channel in real time, when the analog-to-digital converter completes the conversion of the current sample and sets the conversion completion flag, the input capture unit of the coprocessor latches the current value of the global time base counter again to obtain the sampling ready timestamp. By combining the trigger timestamp and the sampling ready timestamp according to the channel number and sampling sequence number of the corresponding sampling data, the two are combined into a double timestamp label. The double timestamp label and the original sampling data are then packaged together into a time-stamped data frame to obtain the original sampling sequence.

4. The multi-source data acquisition method for low-altitude logistics drones according to claim 1, characterized in that, S2 includes: By using the original sampling sequence and its carried dual timestamp information, the start and end times of the current sliding window are read as the time boundaries for unified filtering, and the window time boundary parameters are obtained. By comparing and analyzing the sampling ready timestamp of each sample with the start and end times of the window using the window time boundary parameter, samples whose sampling ready timestamps fall strictly within the window interval are retained, and abnormal samples whose timestamps exceed the window boundaries are removed, thus obtaining the initial screening effective samples for each channel within the current window. By classifying and summarizing the initial screening valid samples by channel, the percentage of valid samples in the current window for each channel relative to the theoretical sampling quantity of that channel is calculated, thus obtaining the completeness statistics for each channel and the set of valid samples for each channel.

5. The multi-source data acquisition method for low-altitude logistics drones according to claim 4, characterized in that, The process involves comparing and analyzing the sampling ready timestamp of each sample with the start and end times of the window using window time boundary parameters. Samples whose sampling ready timestamps strictly fall within the window interval are retained, while abnormal samples whose timestamps exceed the window boundaries are removed. This yields the initial screening of valid samples for each channel within the current window, including: By using the window start time and window end time in the window time boundary parameters, the sampling ready timestamps of all samples in the current window are read point by point to obtain the timestamp value of each sample. The sample set after left boundary filtering is obtained by comparing the timestamp value of each sample with the start time of the window. The sample set filtered by the left boundary is compared with the sampling ready timestamp of each sample in the set and the window end time to determine the initial screening valid samples of each channel in the current window.

6. The multi-source data acquisition method for low-altitude logistics drones according to claim 1, characterized in that, S3 includes: By extracting the trigger timestamp and sampling ready timestamp carried by each valid sample from the set of valid samples in each channel and calculating the difference between the two, the single acquisition delay of each sample is obtained. By calculating the average acquisition delay of all valid samples in the current window based on the single acquisition delay of each sample, the average acquisition delay of each channel in the current window is obtained. By performing time axis reduction on the effective samples of each high-speed dynamic signal channel through the average acquisition delay of each channel, the time coordinate of each sample is reduced from its sampling ready time to its corresponding synchronous trigger time, and the aligned samples with the time coordinates of each channel are unified to the trigger time reference are obtained. By merging and summarizing the aligned samples by channel, signals within the same window point to the same physical time sequence on the time coordinate, thus obtaining a time-aligned multi-source dataset.

7. The multi-source data acquisition method for low-altitude logistics drones according to claim 6, characterized in that, The process involves performing time axis reduction on the effective samples of each high-speed dynamic signal channel using the average acquisition delay of each channel. This process reduces the time coordinate of each sample from its sampling ready time backward to its corresponding synchronization trigger time, resulting in aligned samples within each channel whose time coordinates are unified to the trigger time reference. This includes: By using the average acquisition delay of each channel within the current window, the sampling ready timestamp of each valid sample in the high-speed dynamic signal channel is read to obtain the original time coordinate value of each sample. By subtracting the average acquisition delay of the channel from the original time coordinate value, the reduced time coordinate value is obtained. By replacing the original time coordinate of the sample with the time coordinate value after reduction and keeping the sampling value unchanged, the time axis reduction operation of a single valid sample is completed, and a single aligned sample is obtained. By performing a reduction operation on all valid samples of the high-speed dynamic signal channel one by one, the valid samples of all channels within the same window are aligned to the physical time reference corresponding to the start time of the window on the reduced time coordinate, thus obtaining the aligned samples.

8. The multi-source data acquisition method for low-altitude logistics drones according to claim 1, characterized in that, S4 includes: By extracting the feature time corresponding to the peak value of phase current pulsation in the current channel within the window using time-aligned multi-source datasets, and simultaneously extracting the feature time corresponding to the rising edge of the speed pulse in the speed channel, the current pulsation feature time sequence and the speed pulse feature time sequence are obtained. By using the characteristic time sequence of current pulsation and the characteristic time sequence of rotational speed pulse, the relative phase offset between the two in the time domain of the window is calculated, and the measured time domain phase relationship value of the current window is obtained. The measured time-domain phase relationship value is compared and verified with the theoretical fixed phase value of the corresponding motor pole pair number, and the deviation between the two is calculated to obtain the time consistency deviation indication value. By combining the time consistency deviation indicator value with the integrity statistics of each channel, the data quality of the current window is comprehensively judged and classified, and multi-source data acquisition marking information is obtained.

9. The multi-source data acquisition method for low-altitude logistics drones according to claim 8, characterized in that, The process involves comparing the measured time-domain phase relationship value with the theoretical fixed phase value of the corresponding motor model's pole pairs, calculating the deviation between the two, and obtaining a time consistency deviation indication value, including: The theoretical fixed phase value corresponding to the current number of motor pole pairs is read from the motor parameter configuration table stored on the flight control platform by the measured time-domain phase relationship value of the current window, and the phase verification reference value is obtained. The phase deviation of the current window is obtained by calculating the difference between the measured time-domain phase relationship value and the theoretical reference value. The preset deviation tolerance threshold value is read from the parameter storage area of ​​the flight control platform by the phase deviation amount and the two are compared to determine the time consistency verification result. By analyzing the time consistency verification failure result and the specific numerical direction of the phase deviation, the time lead or lag relationship between the current channel and the speed channel is identified based on the positive or negative sign of the phase deviation. The lead or lag relationship is used as the correction direction indicator for the subsequent window acquisition delay compensation parameters, thereby obtaining the time consistency deviation indication value and the corresponding correction direction information.

10. A multi-source data acquisition device for low-altitude logistics drones, characterized in that, The system includes: The original module is used to allocate different independent sampling periods to multiple types of sensors through the channel configuration table of the flight control platform, drive each sensor to run asynchronously at its own adapted frequency, and perform timestamp analysis and marking on each sampled data to obtain the original sampling sequence; The time analysis module is used to compare and analyze the sampling ready timestamp of each sample with the window start and end times using the original sampling sequence and its carried dual timestamp information, with the sliding window boundary as a unified screening criterion, to obtain the effective sample set and its completeness statistics. The time alignment module is used to perform time axis reduction on the effective samples of each high-speed dynamic signal channel according to their respective average acquisition delay, thereby obtaining a time-aligned multi-source dataset. The acquisition and marking module is used to extract the current pulsation feature time and speed pulse feature time within the window by aligning the multi-source dataset with time, compare and verify the time-domain phase relationship between the two with the theoretical fixed phase value of the corresponding motor model, and perform quality analysis and marking to obtain multi-source data acquisition and marking information.

Images