An underwater hard sweeping bed shallow point touch automatic monitoring method and device

Abstract

The application discloses an underwater hard sweeping bed shallow point touch automatic monitoring method, and steps are as follows: S1, real-time data is acquired; S2, a frequency band range that can be analyzed is set through a short time window and a sensor sampling rate, high frequency noise and a periodic signal with relatively strong energy are removed to obtain time domain convolution coefficients of frequency domain filtering, and convolution processing is directly carried out on sampled posture data; S3, posture data of a filtered sweeping bed frame vertical rod is convolved with a detection convolution kernel function, local maximum values are found, mean values and standard deviations of the local maximum values are analyzed, and shallow point touch signals are recognized according to a 3 times standard deviation principle; and S4, weight coefficients are constructed according to multiple time scale parameters, and scoring on detection results of suspected touch signals at each moment is realized. The application further discloses an underwater hard sweeping bed shallow point touch automatic monitoring device. The application can realize swing automatic monitoring and recognition of two types of shallow point touch events of a sweeping bed frame, and can be widely applied in the field of navigation safety.

Description

Technical Field

[0001] This invention relates to the design of waterways and related fields such as underwater dredging, reef blasting, and obstacle removal, and in particular to an automatic monitoring method and device for shallow contact of underwater rigid sweeping beds. Background Technology

[0002] Rigid sweepers, also known as fixed-depth sweepers, are typically installed on shallow-draft vessels. The sweeper frame consists of a swingable vertical bar and a horizontal bar submerged underwater. Before sweeping operations, the sweeper frame must be adjusted so that the horizontal bar is lowered horizontally to the designed channel depth. Then, the vessel is moved to perform the sweeping operation. When an obstacle is encountered that exceeds the designed channel depth, it will collide with the underwater horizontal bar, causing the vertical bar to swing. This allows the location of obstacles that have not reached the designed channel depth to be identified. Therefore, rigid sweepers are an essential means of quality inspection in channel improvement projects. Their sweeper bars are rigid, and the depth of immersion is limited. Although they cannot directly measure the depth of underwater topographic points, they can visually reflect whether the underwater topographic points meet the bottom elevation design requirements.

[0003] At present, the equipment and operation process of rigid sweeping machines are still relatively traditional, and there is a lack of rapid acquisition and informatization and digital processing of information such as water depth and status of rigid sweeping equipment. The swaying of the sweeping machine frame can be monitored by adding sensors such as pressure, attitude or inclinometer, and it can be determined whether the crossbar of the sweeping machine frame has bottomed out. The ideal swaying situation of the crossbar of the sweeping machine frame touching the bottom at a shallow point is mainly divided into two types of characteristics: (1) The bottom crossbar of the sweeping machine frame fails to fall off after touching the shallow point, the vertical bar connecting the sweeping machine frame swings at a large angle, or even deforms, and the sweeping machine operation is forced to stop. The sweeping machine frame attitude monitoring data shows a large step signal change; (2) The underwater shallow point is within the critical range of the design elevation. The crossbar of the sweeping machine frame scrapes against the shallow point, causing a relatively large swaying, and falls off and returns to its position in a short time. The sweeping machine frame attitude monitoring data shows a large peak signal.

[0004] However, the actual sweeping operation involves a complex swinging process of the sweeping frame. The sweeping frame is subjected to water flow impact, which generates rapid "vortex street" oscillations. Numerous disturbances, such as the sweeping boat's own attitude changes during navigation, short-term attitude changes caused by surrounding navigation interference, and changes in the frequency and amplitude of "vortex street" oscillations caused by changes in boat speed and direction of water flow, make it very difficult to stably and accurately capture shallow point contact events of the sweeping frame. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the aforementioned background technology and provide an automatic monitoring method and device for shallow point contact of underwater rigid sweeping beds. This method fully considers the interference of factors such as "vortex street" oscillation caused by water flow impact under actual working conditions, changes in the attitude of the sweeping bed boat and other vessels, and changes in the "vortex street" oscillation characteristics caused by changes in boat speed and direction. Through the combined application of short-time frequency domain analysis and dynamic filtering, multi-scale comprehensive judgment of shallow point contact step signals, and multi-scale comprehensive judgment of shallow point contact peak signals, the method can realize automatic monitoring and identification of the swaying of two types of shallow point contact events of the sweeping bed frame.

[0006] This invention provides an automatic monitoring method for shallow point contact of an underwater rigid sweeping bed, comprising the following steps: S1, calculating the elevation of the crossbar in real time using a pressure sensor, collecting the swing posture data of the vertical bar of the sweeping bed frame in real time using an inclinometer and processing and analyzing the data in real time, and providing real-time position and direction of travel data using a satellite positioning device; S2, using a short-time window t and sensor sampling rate f 0 sets the range of frequency bands that can be resolved. f [ f 0 / 2,2 / t [S1] Remove high-frequency noise and high-energy periodic signals to obtain the time-domain convolution coefficients for frequency domain filtering. Convolve these time-domain convolution coefficients directly with the sampled pose data to obtain the filtered pose data. S2] Construct a detection convolution kernel function with scale resolution. After each update of the sampled data, the filtered pose data for each time step... t The posture data of the nearby sweeping bed frame vertical rods are convolved with the detection convolution kernel function to find the local maxima in the convolution sequence. The mean and standard deviation of these local maxima are analyzed, and shallow touch signals are identified according to the principle of 3 times the standard deviation. S4. Weight coefficients are constructed based on multiple preset time scale parameters. If a suspected touch signal is detected at each scale at the same time, it is counted as 1. If no suspected touch signal is detected, it is counted as 0, so as to score the detection results of suspected touch signals at each time.

[0007] In the above technical solution, the specific process of step S2 is as follows: S21, Select a suitable spectrum analysis time window. t and sensor sampling rate f 0, based on the attitude sampling data every t A time window is set for a duration of 2 seconds. t S22. Short-time spectrum analysis; S23. Set a frequency domain filter template, perform band-stop processing on periodic signals with strong energy on the spectrum graph, and perform high-frequency blocking processing on noise. Then, perform inverse Fourier transform on the frequency domain filter template to form a time domain signal convolution template; S24. In the dynamic sampling process of the vertical rod attitude data of the sweeping bed frame, let the current sampling time be... t 0, will [ t0- t , t The result is obtained by multiplying the data of length 0 by the time-domain convolution coefficients of the time-domain signal convolution template and summing the results. t 0- t Attitude data after filtering at time / 2.

[0008] In the above technical solution, in step S22, the process of band-stopping the periodic signal with strong energy on the spectrum and blocking the high frequency of noise involves assigning zeros to both the periodic signal frequency band and the high frequency band of the frequency domain filter template, and assigning 1s to the remaining frequency bands other than the periodic signal frequency band and the high frequency band.

[0009] In the above technical solution, the mathematical expressions for steps S21 and S22 are as follows:

[0010] ,in, E ( f () represents the energy at each frequency. F ( f ) represents a frequency domain filter that removes high-frequency noise and a periodic signal with high energy, respectively. t ) represents the time-domain convolution coefficients of the frequency-domain filter.

[0011] In the above technical solution, in step S3, the shallow touch signal is a shallow touch step signal, and the specific process of the shallow touch step signal detection method is as follows: S31, preset multiple time scale parameters. s According to each time scale parameter s Generate step signal detection convolution kernel function; S32, for each scale parameter s And the corresponding step signal detection convolution kernel function, after each update of the sampled data, the filtered data at each time step is... t exist[ t -3 s , t +3 s The attitude data of the nearby sweeping bed frame vertical rod are multiplied by each coefficient of the step signal detection convolution kernel function and summed to obtain... t S33. Determine whether the above convolution result is a local maximum, and calculate the mean ave0 and standard deviation std0 of the most recently determined m local maxima. If the current detection result is a local maximum and the absolute value obtained by subtracting the mean ave0 from the mean is greater than 3 * standard deviation std0, then mark the time as a suspected step change point.

[0012] In the above technical solution, the step signal detection convolution kernel function is as follows: , among which, tanh( t) is the tangent function, s This is a time-scale parameter used to adjust tanh( t The time width of the function value transition; t For step signal detection, the independent variable of the convolution kernel function is 2 and the denominator (6). s +1) is used to calculate the regularization coefficient.

[0013] In the above technical solution, in step S3, the shallow touch signal is a shallow touch peak signal, and the specific process of the shallow touch peak signal detection method is as follows: S31, preset multiple time scale parameters. s According to each time scale parameter s Generate peak signal detection convolution kernel function; S32, for each scale parameter s And the corresponding peak signal detection convolution kernel function, after each update of the sampled data, the filtered data at each time step is... t exist[ t -3 s , t +3 s The attitude data of the nearby sweeping bed frame vertical rods are multiplied by each coefficient of the peak signal detection convolution kernel function and summed to obtain... t S33. Determine whether the above convolution result is a local maximum, and calculate the mean ave0 and standard deviation std0 of the m most recently determined local maxima. If the current detection result is a local maximum and the absolute value obtained by subtracting the mean ave0 from the mean is greater than 3 * standard deviation std0, then mark the time as a suspected peak abrupt change point.

[0014] In the above technical solution, the peak signal detection convolution kernel function is as follows:

[0015] ,in, s For time scale parameters; t is the independent variable of the convolution kernel function for peak signal detection.

[0016] In the above technical solution, in step S4, the scoring of the suspected touch signal detection results at each moment is performed according to the following formula: ,in, s For time scale parameters; t The independent variable is the result of the touch signal detection.

[0017] The present invention also provides an automatic monitoring device for shallow contact of an underwater rigid sweeping bed, which has a computer program that can execute an automatic monitoring method for shallow contact of an underwater rigid sweeping bed.

[0018] The present invention provides an automatic monitoring method and device for shallow point contact of underwater rigid sweeping beds, which has the following beneficial effects:

[0019] Compared to the traditional observation, judgment, and recording mode of rigid sweeping machines, this invention utilizes digital monitoring technology for the rigid sweeping machine operation process. It fully considers the signal characteristics of the two most common shallow touch processes in rigid sweeping machine operations and designs a multi-scale detection convolution kernel function and a real-time detection algorithm. This can accurately, stably, and efficiently detect and identify shallow touch events in the rigid sweeping machine process, thereby realizing the digitalization, automation, and intelligence of rigid sweeping machine operations. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall process of the underwater rigid sweeping bed shallow point contact automatic monitoring method of the present invention;

[0021] Figure 2 This is a schematic diagram of the hardware environment upon which step one of the underwater rigid sweeping bed shallow point contact automatic monitoring method of the present invention depends;

[0022] Figure 3 This is a flowchart illustrating step two of the underwater rigid sweeping bed shallow point contact automatic monitoring method of the present invention;

[0023] Figure 4 This is a schematic diagram of the short-time spectrum analysis and filter update process in step two of the underwater rigid sweeping bed shallow point contact automatic monitoring method of the present invention;

[0024] Figure 5 This is a schematic diagram of the short-time spectrum analysis and dynamic filtering results before and after step two in the underwater rigid sweeping bed shallow point contact automatic monitoring method of the present invention.

[0025] Figure 6 This is a schematic diagram illustrating the coefficients of the multi-scale step signal detection convolution kernel function constructed in the shallow point touch step signal detection method of step three in the underwater rigid sweeping bed automatic monitoring method of the present invention.

[0026] Figure 7 This is a flowchart illustrating the shallow touch step signal detection method (including the multi-scale judgment process of the step signal) in step three of the underwater rigid sweeping bed shallow touch automatic monitoring method of the present invention.

[0027] Figure 8 This is a schematic diagram illustrating the coefficients of the multi-scale peak signal detection convolution kernel function constructed in the shallow point touch peak signal detection method of step three in the underwater rigid sweeping bed automatic monitoring method of the present invention.

[0028] Figure 9 This is a flowchart illustrating the shallow touch peak signal detection method (including the multi-scale judgment process of peak signal) in step three of the underwater rigid sweeping bed shallow touch automatic monitoring method of the present invention.

[0029] Figure 10 This is a schematic diagram of the underwater rigid sweeping bed shallow point contact automatic monitoring device of the present invention. Detailed Implementation

[0030] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments, but these embodiments should not be construed as limiting the present invention.

[0031] Technical Principles of the Invention

[0032] Based on the operational characteristics of rigid sweeping beds, this invention designs an automated monitoring system for rigid sweeping beds. It fully considers the interference from factors such as vortex street oscillations caused by water flow impact under actual working conditions, changes in the attitude of the sweeping bed boat and other vessels, and changes in the characteristics of vortex street oscillations caused by changes in boat speed and direction. Through the combined application of short-time frequency domain analysis and dynamic filtering, multi-scale comprehensive judgment of shallow-point touch step signals, and multi-scale comprehensive judgment of shallow-point touch peak signals, it can achieve automatic monitoring and identification of the swaying of two types of shallow-point touch events on the sweeping bed frame.

[0033] Specific technical solutions

[0034] See Figure 1 The technical solution of the underwater rigid sweeping bed shallow point contact automatic detection method of the present invention includes the following contents:

[0035] Step 1: The elevation of the crossbar is calculated in real time using a pressure sensor; the inclinometer collects and processes the swing posture data of the vertical bar of the sweeping bed frame in real time; and the satellite positioning equipment provides real-time position and direction of travel data. See below for details. Figure 2 :

[0036] 1. Construction of a rigid sweeping bed monitoring system

[0037] To achieve automatic detection of shallow contact in rigid scanning beds, it is necessary to build a necessary hardware system, which mainly includes pressure sensors, inclinometers, satellite positioning modules, host computers for receiving and analyzing various types of data, and output display terminals.

[0038] 1. The pressure sensor can monitor the depth of the crossbar of the sweeping bed frame relative to the water surface in real time. Combined with the water surface elevation, the lowering elevation of the crossbar underwater can be determined.

[0039] 2. Use a tilt meter sensor with a sampling rate of ≥20Hz to monitor the swing posture of the sweeping bed frame vertical rod, process and analyze the data in real time, and determine whether the aforementioned two types of shallow touch events have occurred.

[0040] 3. The satellite positioning module provides real-time location and direction of travel data. Combined with the relative installation of the sweeping frame and the positioning receiver, it calculates the geographical coordinates of both ends of the bottom crossbar of the sweeping frame in real time, thereby determining the area swept by the crossbar of the sweeping frame.

[0041] Step 2: Using a short window t and sensor sampling rate f 0 sets the range of frequency bands that can be resolved. f [ f 0 / 2,2 / t The process involves removing high-frequency noise and high-energy periodic signals to obtain the time-domain convolution coefficients for frequency domain filtering. These time-domain convolution coefficients are then directly convolved with the sampled attitude data to obtain the filtered attitude data. In one or more embodiments, an example of short-time spectrum analysis and filter update is as follows: sampling rate 20Hz, short-time analysis window 6s, resolution frequency band [0.33, 10]Hz, with relatively stable periodic oscillations I and II. Frequency domain band-stop I, band-stop II, and high-stop filter coefficients are set respectively, and converted to the time domain to form time-domain convolution template coefficients. The specific content is as follows: Figure 3 As shown:

[0042] 1. Short-time spectrum analysis of the vertical posture data of the bed frame

[0043] like Figure 4 As shown, select an appropriate time window for spectrum analysis. t and sensor sampling rate f 0, based on the attitude sampling data every t A time window is set for a duration of 2 seconds. t Short-time spectral analysis.

[0044] 2. Construction of dynamic filters

[0045] like Figure 5 As shown, the swing posture data of the vertical bar of the sweeping bed frame. S ( t The vortex oscillation is the result of the combined effects of multiple factors, caused by the impact of water flow. f vortex Swaying caused by the swaying of the sweeper itself and other shipboard disturbances f vessel These signals can be considered stable periodic signals over a short period of time, and the observation noise often exhibits high-frequency characteristics. f noise Shallow touch causes a step change or abrupt change in the vertical rod's attitude, which manifests as low-frequency characteristics. L ( f Therefore, high-frequency noise and periodic interference signals can be attenuated through frequency domain filtering, and the attenuated attitude data is as follows: S '( t However, frequency domain filtering based on Fourier transform lacks time-sensitivity. Therefore, Short-Time Fourier Transform (FFT) and Inverse Fourier Transform (RFFT) are introduced, along with short-time window filtering. t and sensor sampling rate fThe 0s together determine the range of frequency bands that can be resolved. f [ f 0 / 2,2 / t The energy of each frequency is expressed as: E ( f For high-frequency noise and periodic signals with strong energy, frequency domain filters can be set separately. F ( f The removal of ) can be described by the following system of equations:

[0046] (1),

[0047] In the above formula, filter( t These are the time-domain convolution coefficients for frequency-domain filtering;

[0048] In one or more embodiments, examples of short-time spectrum analysis and results before and after dynamic filtering are as follows: the short-time window is 2s, the frequency resolution range is [1,10]Hz, some periodic changes and high-frequency noise are suppressed after filtering, but there is still a signal with an approximate period of 0.5Hz. To resolve this signal, the window length needs to be increased, which reduces the timeliness of the analysis.

[0049] 3. Temporal Convolution Template Coefficient Analysis

[0050] By directly convolving the temporal convolution template coefficients with the sampled pose data, the filtered pose data can be obtained.

[0051] Step 3: Construct a detection convolution kernel function with scale resolution. After each update of the sampled data, the filtered kernel function at each time step is used. t The posture data of the nearby sweeping bed frame vertical rods were convolved with the detection convolution kernel function to find local maxima in the convolution sequence. The mean and standard deviation of these local maxima were analyzed, and shallow touch signals were identified according to the principle of 3 times the standard deviation. The specific details are as follows:

[0052] 1. Multi-scale detection method for shallow touch step signal

[0053] The bottom crossbar of the sweeping bed frame failed to detach after contacting the shallow point, forcing the sweeping operation to stop. The sweeping bed frame attitude monitoring data showed a large step signal change. The step signal had a clear first derivative maximum. However, considering that the step rise was not instantaneous but a process lasting for a very short time, the step signal generated by the shallow point contact had multi-scale characteristics. Based on the characteristics of the tanh function, this invention constructs a quasi-first derivative convolution kernel function with scale resolution capability. This is used to construct the convolution kernel function, namely the step signal detection convolution kernel function, as shown in formula (2):

[0054] (2),

[0055] In the above formula, tanh( t The function takes values ​​in the range (-1, 1). s This is a time scale parameter used to adjust the time span of the function value transitioning from -1 to 1, such as... Figure 6 As shown; kernel function independent variable t The value range is set to [-3] s ,3 s During real-time detection, the calculation process has 3... s The lag in duration; coefficient 2 and denominator (6) s +1) is required for computational regularization.

[0056] The steps for multi-scale detection of shallow touch step signals are as follows: Figure 7 As shown, the scanning bed posture monitoring data is convolved with the above template (step signal detection convolution kernel function). The convolution result of the step signal is more easily amplified, and the convolution result of the shallow touch step signal will show abnormal maxima. Find the local maxima in the sequence, analyze the mean and standard deviation of these local maxima, and identify the step signal caused by shallow touch according to the 3-times-standard-deviation principle.

[0057] In addition, during real-time data sampling and detection, at the same time t At point 0, the timeliness of detection varies depending on the scale. s The smaller the value, the more timely the judgment, but the lower the reliability. s The larger the size, the worse the timeliness, but the stronger the smoothing ability on both sides of the template, thus improving reliability. Comprehensive judgment at multiple scales will greatly improve the reliability of shallow touch step signal detection.

[0058] 2. Multi-scale detection method for shallow touch peak signals

[0059] The underwater shallow point is within the critical range of the design elevation. The crossbar of the sweeping bed frame scrapes against the shallow point, causing relatively large swings, and then falls back into place within a short time. The sweeping bed frame attitude monitoring data shows a large peak signal. This signal also has a time width, and multi-scale factors need to be considered when detecting this signal. Based on the characteristics of the second-order Gaussian function, this invention constructs a peak signal detection convolution kernel function with scale resolution capability, as shown in formula (3):

[0060] (3),

[0061] In the above formula, s This is a time scale parameter used to adjust the time width of the peak portion in the middle of the template, such as... Figure 8 As shown; kernel function independent variable t The value range is set to [-3] s ,3 sDuring real-time detection, the calculation process also has 3... s The time lag. Apart from the difference in kernel function mentioned above, the other steps of the multi-scale shallow touch peak signal detection method are the same as those of the shallow touch step signal detection method, such as... Figure 9 As shown.

[0062] Step 4: Construct weighting coefficients based on multiple preset time scale parameters. If a suspected touch signal is detected at any scale at the same time, count it as 1; if no suspected touch signal is detected, count it as 0. This achieves scoring of the suspected touch signal detection results at each time point (multi-scale judgment of shallow touch step signals and shallow touch peak signals is performed separately at each scale). Figure 7 and Figure 9 (This was disclosed in the document), and the specific process is as follows:

[0063] This invention is based on multiple preset time scale parameters [ s 0< s 1< s 2< s 3<…] Construct weight coefficients. If a suspected step signal is detected at each scale at the same time, it is counted as 1. If no signal is detected, it is counted as 0. This allows for scoring of the suspected step signal detection results at each time point, as shown in formula (4):

[0064] (4),

[0065] in, s For time scale parameters; t The independent variable is the result of the touch signal detection.

[0066] See Figure 10 The present invention relates to an automatic detection device for shallow point contact of an underwater rigid sweeping bed, comprising the following parts:

[0067] Data acquisition module: The pressure sensor calculates the elevation of the crossbar in real time, the inclinometer collects the swing posture data of the vertical bar of the sweeping bed frame in real time and processes and analyzes the data in real time, and the satellite positioning equipment provides real-time position and direction of travel data;

[0068] Spectrum analysis and filtering module: using a short-time window t and sensor sampling rate f 0 sets the range of frequency bands that can be resolved. f [ f 0 / 2,2 / t The high-frequency noise and high-energy periodic signals are removed to obtain the time-domain convolution coefficients of the frequency domain filter. The time-domain convolution coefficients are then directly convolved with the sampled attitude data to obtain the filtered attitude data.

[0069] Shallow touch signal detection and recognition module: Constructs a detection convolution kernel function with scale resolution capability. After each update of the sampled data, the filtered data at each time step is processed. t The posture data of the nearby sweeping bed frame vertical rods are convolved with the detection convolution kernel function to find the local maxima in the convolution sequence. The mean and standard deviation of these local maxima are analyzed, and shallow touch signals are identified according to the principle of 3 times the standard deviation.

[0070] The shallow touch signal multi-scale judgment module constructs weight coefficients based on multiple preset time scale parameters. If a suspected touch signal is detected at each scale at the same time, it is counted as 1; if no suspected touch signal is detected, it is counted as 0, thus scoring the detection results of suspected touch signals at each time.

[0071] Example

[0072] This invention constructs a rigid sweeping bed detection system to achieve automatic detection of shallow touches on the sweeping bed frame. The following is a detailed description of a specific implementation case. The implementation process mainly includes: installation of the rigid sweeping bed monitoring hardware system, short-time spectrum analysis and real-time filtering of sweeping bed frame attitude data, detection of suspected step and peak signals at different scales, and multi-scale judgment of step and peak signals.

[0073] (1) Installation of sensors in the rigid sweeping bed monitoring system

[0074] The monitoring system for rigid sweeping beds involves sensors including pressure sensors, tilt sensors, and positioning and orientation receivers, such as... Figure 2 As shown. The pressure sensor is located near the bottom crossbar of the sweeper frame. Based on the pressure value and other fixed parameters, it calculates the elevation of the crossbar in real time, thus guiding the placement of the sweeper frame during operation. The inclinometer is fixed on the vertical bar of the sweeper frame, and the rotation axis of the inclinometer should be parallel to the swing axis of the vertical bar. Positioning and orientation employ a dual-satellite receiver scheme, with the two receivers deployed along the bow and stern of the ship. The relative positional relationship between the main positioning receiver and each vertical bar of the sweeper frame is measured, and the coordinates of the vertical bars of the sweeper frame are calculated in real time based on the receiver positioning and orientation data.

[0075] (2) Short-time spectrum analysis and real-time filtering of bed frame attitude data

[0076] ① Select an appropriate time window for spectrum analysis t , t It is advisable to set it to 2~5s, which corresponds to a minimum resolvable frequency of 1~0.2Hz, and a resolvable high frequency of half the system sampling rate.

[0077] Based on the attitude sampling data every t A time window is set for a duration of 2 seconds. t Short-time spectrum analysis, such as Figure 4 As shown.

[0078] ② Set up a frequency domain filtering template, perform band-stop processing on periodic signals with strong energy on the spectrum graph, and perform high-frequency blocking processing on noise. That is, assign zeros to the periodic frequency band and high-frequency band of the frequency domain filtering template, and assign 1s to the remaining frequency band. Then perform inverse Fourier transform on the frequency domain filtering template to form a time domain signal convolution template.

[0079] ③ During the dynamic sampling process of the bed frame vertical pole attitude data, the current sampling time is t 0, will [ t 0- t , t The result is obtained by multiplying the data of length 0 with the coefficients of the temporal convolution template and summing the results. t 0- t Filtered data at time / 2, such as Figure 5 As shown.

[0080] (3) Detection of suspected step and peak signals at different scales

[0081] Figure 7 and Figure 9 The basic process of multi-scale detection of step signals and peak signals is given respectively.

[0082] First, preset 3 to 5 time scale parameters. s [ s 0 = 0.2, s 1 = 0.4, s 2 = 0.6, s 3=0.8,…], according to each s Generate step signal detection convolution kernel and peak signal detection convolution kernel according to formulas (2) and (3) respectively.

[0083] Secondly, for each scale parameter s And the corresponding convolution kernel, after each update of the sampled data, the filtered scanning bed frame vertical rod attitude monitoring data at each moment. t nearby[ t -3 s , t +3 s The data is multiplied by the coefficients of each of the two convolution kernels and then summed to obtain... t The convolution result at time step 1.

[0084] Then, determine whether the above convolution result is a local maximum, and calculate the mean ave0 and standard deviation std0 of the 10 most recently determined local maxima. If the current detection result is a local maximum, and the absolute value after subtracting ave0 from it is greater than 3*std0, then mark that moment as a suspected step change point or peak abrupt change point.

[0085] (4) Multi-scale judgment of step and peak signals

[0086] According to formula (4), the suspected signals detected at each sampling time are scored, with a score range of [0~1] points. A score of 0.7 or above is set as a reliable shallow touch event.

[0087] Technical effect

[0088] Compared with the prior art, the present invention has the following technical effects:

[0089] Compared to the traditional observation, judgment, and recording mode of rigid sweeping machines, this invention utilizes digital monitoring technology for the rigid sweeping machine operation process. It fully considers the signal characteristics of the two most common shallow touch processes in rigid sweeping machine operations and specifically designs multi-scale convolution kernels and real-time detection algorithms. This can accurately, stably, and efficiently detect and identify shallow touch events in the rigid sweeping machine process, thereby realizing the digitalization, automation, and intelligence of rigid sweeping machine operations.

[0090] 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.

[0091] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. An automatic detection method for shallow point contact of an underwater rigid sweeping bed, characterized in that: Includes the following steps: S1. The elevation of the crossbar is calculated in real time by the pressure sensor, the inclinometer collects the swing posture data of the vertical bar of the sweeping bed frame in real time and processes and analyzes the data in real time, and the satellite positioning equipment provides real-time position and direction of travel data. S2, via short window τ and sensor sampling rate f 0 sets the range of frequency bands that can be resolved. f [ f 0 / 2,2 / τ The high-frequency noise and high-energy periodic signals are removed to obtain the time-domain convolution coefficients of the frequency domain filter. The time-domain convolution coefficients are then directly convolved with the sampled attitude data to obtain the filtered attitude data. S3. Construct a detection convolution kernel function with scale resolution. After each update of the sampled data, the filtered kernel function at each time step is used. t The posture data of the nearby sweeping bed frame vertical rods are convolved with the detection convolution kernel function to find local maxima in the convolution sequence. The mean and standard deviation of these local maxima are analyzed, and shallow touch signals are identified according to the 3-standard-deviation principle. The shallow touch signal is a shallow touch step signal. The specific process of the shallow touch step signal detection method is as follows: S31, Preset multiple time scale parameters σ According to each time scale parameter σ Generate step signal detection convolution kernel function; S32, For each scale parameter σ And the corresponding step signal detection convolution kernel function, after each update of the sampled data, the filtered data at each time step is... t exist[ t -3 σ , t +3 σ The attitude data of the nearby sweeping bed frame vertical rod are multiplied by each coefficient of the step signal detection convolution kernel function and summed to obtain... t The convolution result at time step, the convolution kernel function for the step signal detection is as follows: ， Among them, tanh( t ) is the tangent function, σ The time scale parameter is used to adjust tanh( t The time width of the function value transition; t For step signal detection, the independent variable of the convolution kernel function is 2 and the denominator (6). σ +1) is used to calculate the regularization coefficient; S33. Determine whether the above convolution result is a local maximum, and calculate the mean ave0 and standard deviation std0 of the most recently determined m local maxima. If the current detection result is a local maximum and the absolute value obtained by subtracting the mean ave0 from the mean is greater than 3 * standard deviation std0, then mark the moment as a suspected step change point. or, The shallow touch signal is a shallow touch peak signal, and the specific process of the shallow touch peak signal detection method is as follows: S31, Preset multiple time scale parameters σ According to each time scale parameter σ Generate peak signal detection convolution kernel function; S32, For each scale parameter σ And the corresponding peak signal detection convolution kernel function, after each update of the sampled data, the filtered data at each time step is... t exist[ t -3 σ , t +3 σ The attitude data of the nearby sweeping bed frame vertical rods are multiplied by each coefficient of the peak signal detection convolution kernel function and summed to obtain... t The convolution result at time step 1, the peak signal detection convolution kernel function is as follows: ， in, σ For time scale parameters; t The independent variable of the convolution kernel function for peak signal detection; S33. Determine whether the above convolution result is a local maximum, and calculate the mean ave0 and standard deviation std0 of the most recently determined m local maxima. If the current detection result is a local maximum and the absolute value obtained by subtracting the mean ave0 from the mean is greater than 3 * standard deviation std0, then mark that moment as a suspected peak abrupt change point. S4. Construct weighting coefficients based on multiple preset time scale parameters. If a suspected touch signal is detected at each scale at the same time, count it as 1. If no suspected touch signal is detected, count it as 0. This achieves scoring of the suspected touch signal detection results at each time.

2. The underwater rigid sweeping bed shallow point contact automatic monitoring method according to claim 1, characterized in that: The specific process of step S2 is as follows: S21. Select an appropriate spectrum analysis time window. τ and sensor sampling rate f 0, based on the attitude sampling data every τ A time window is set for a duration of 2 seconds. τ Short-time spectrum analysis; S22. Set up a frequency domain filtering template, perform band-stop processing on periodic signals with strong energy on the spectrum, and perform high-frequency blocking processing on noise. Then, perform an inverse Fourier transform on the frequency domain filtering template to form a time domain signal convolution template. S23. During the dynamic sampling process of the bed frame vertical rod attitude data, let the current sampling time be... t 0, will [ t 0- τ , t The result is obtained by multiplying the data of length 0 by the time-domain convolution coefficients of the time-domain signal convolution template and summing the results. t 0- τ Attitude data after filtering at time / 2.

3. The underwater rigid sweeping bed shallow point contact automatic monitoring method according to claim 2, characterized in that: In step S22, the process of band-stopping the periodic signal with strong energy on the spectrum and blocking the high frequency of noise involves setting zeros to the periodic signal frequency band and the high frequency band of the frequency domain filter template, and setting 1s to the remaining frequency bands other than the periodic signal frequency band and the high frequency band.

4. The underwater rigid sweeping bed shallow point contact automatic monitoring method according to claim 3, characterized in that: The mathematical expressions for steps S21 and S22 are as follows: ， in, E ( f () represents the energy at each frequency. F ( f ) represents a frequency domain filter that removes high-frequency noise and a periodic signal with high energy, respectively. t ) represents the time-domain convolution coefficients of the frequency-domain filter.

5. The underwater rigid sweeping bed shallow point contact automatic monitoring method according to claim 4, characterized in that: In step S4, the scoring of the suspected touch signal detection results at each time point is performed using the following formula: ， in, σ For time scale parameters; t The independent variable is the result of the touch signal detection.

6. An automatic monitoring device for shallow point contact of an underwater rigid sweeping bed, comprising a computer program, characterized in that: The computer program is capable of executing the underwater rigid sweeping bed shallow point contact automatic detection method as described in any one of claims 1 to 5.

Images