An adaptive control method and system for a hydraulic drive guniting robot

Abstract

The present application relates to the technical field of hydraulic drive industrial control, and particularly discloses a self-adaptive control method and system for a hydraulic drive guniting manipulator, which separates a composite signal into a low-frequency characteristic signal reflecting load change and a high-frequency characteristic signal reflecting pressure impact by real-time acquisition of pressure fluctuation signals, flow change signals and mechanical arm displacement signals of a hydraulic system, combined with frequency domain decomposition processing technology; for the low-frequency signal, a dynamic weight adjustment strategy is adopted to generate a load compensation control quantity; for the high-frequency signal, a phase compensation process with adjustable differential order is used to suppress pressure oscillation; finally, the two control characteristic values are input into a drive control model of the guniting manipulator to generate an adjustment instruction of a hydraulic actuator.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of hydraulic drive industrial control, and in particular to a self-adaptive control method and system for a hydraulic drive shotcrete manipulator. BACKGROUND

[0002] In the field of modern industrial control automation, hydraulic drive manipulators are widely used in heavy load, high precision or complex trajectory control scenarios such as building shotcrete, mine support, metallurgical casting and other harsh working conditions. Traditional hydraulic control systems mostly use PID control or feedforward compensation strategies, relying on fixed parameter models, which are difficult to adapt to complex dynamic characteristics such as load mutation, external disturbance and system nonlinearity. Especially in shotcrete operation, the hydraulic system is frequently started and stopped, and the flow and pressure fluctuate dramatically, which can easily cause system oscillation, response lag and control precision decline. In addition, the manipulator is affected by the coupling of inertia force, friction force and external load during movement, making it difficult for the industrial control model to accurately describe its dynamic behavior.

[0003] The prior art has the following disadvantages:

[0004] The present application aims to solve the technical problem that traditional industrial control methods are difficult to achieve multi-frequency band feature signal collaborative response and adaptive compensation under the coupling effect of complex dynamic load and high-frequency pressure impact for a hydraulic drive shotcrete manipulator. The prior art usually processes pressure, flow and displacement signals uniformly, ignoring the physical meaning represented by different frequency band signals and their differentiated influence on system control, resulting in response delay, overcompensation or undercompensation of the control strategy when facing load mutation and impact disturbance.

[0005] The present application separates the composite signal into low-frequency feature signals reflecting load changes and high-frequency feature signals reflecting pressure impact through frequency domain decomposition processing technology, and respectively uses dynamic weight adjustment and differential order adjustable phase compensation mechanism to realize differentiated processing and fusion control of multi-frequency band features. This method not only improves the response ability and industrial control precision of the system to complex working conditions, but also effectively suppresses the pressure oscillation of the hydraulic system, filling the technical gap in multi-frequency band feature collaborative control in the current industrial control field. SUMMARY

[0006] The present application aims to provide a self-adaptive control method and system for a hydraulic drive shotcrete manipulator to solve the problems in the above background.

[0007] The object of the present application can be achieved by the following technical solutions:

[0008] A self-adaptive control method for a hydraulic drive shotcrete manipulator, comprising the following steps:

[0009] S1: Collecting pressure fluctuation signal, flow change signal and mechanical arm displacement signal of the hydraulic system in real time;

[0010] S2: Frequency domain decomposition processing is performed on the collected signals, and low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals reflecting pressure impact are separated;

[0011] S3: Dynamic weight adjustment processing is performed on the low-frequency characteristic signals to generate first control characteristic values for compensating load changes;

[0012] S4: The high-frequency characteristic signals are subjected to phase compensation processing with adjustable differentiation order to generate second control characteristic values for suppressing pressure oscillation;

[0013] S5: The first control characteristic values and the second control characteristic values are input into the driving control model of the shotcrete mechanical hand to generate the adjustment instructions of the hydraulic actuator.

[0014] As a further scheme of the application: the frequency domain decomposition processing comprises the following specific steps:

[0015] A signal separation device based on a multi-stage filter is constructed, a Butterworth filter with adjustable cutoff frequency is used in the first stage to extract 0-100Hz frequency band signals as low-frequency characteristic signals, and a comb filter group is used in the second stage to extract harmonic components of specific frequency points greater than 1kHz as high-frequency characteristic signals;

[0016] The low-frequency characteristic signals are subjected to sliding window normalization processing, the window length is dynamically adjusted according to the movement speed of the mechanical arm, and the normalized signals are sent to the subsequent processing module;

[0017] The high-frequency characteristic signals are subjected to amplitude reconstruction based on time domain envelope detection, the time-varying characteristics of the shock waveforms are retained, and the interference of high-frequency carrier components is eliminated.

[0018] As a further scheme of the application: the frequency domain decomposition processing further comprises the following optimization steps:

[0019] Wavelet packet transform is used to decompose the original signals by 16 layers, the 3rd-5th layer detail coefficients are selected to reconstruct the low-frequency characteristic signals, and the 10th-12th layer detail coefficients are selected to reconstruct the high-frequency characteristic signals;

[0020] The reconstructed high-frequency characteristic signals are subjected to adaptive threshold denoising processing, and the threshold size is dynamically adjusted according to the signal-to-noise ratio under the current working condition;

[0021] The cross-correlation function of the low-frequency characteristic signals and the high-frequency characteristic signals is established, and when the correlation coefficient exceeds the set threshold, the signal compensation mechanism is started to correct the decomposition results.

[0022] As a further scheme of the present application, the dynamic weight adjustment process comprises the following specific steps:

[0023] A feature extraction network with three levels of hidden layers is constructed, the first level network extracts the pressure-flow coupling features of the hydraulic system, the second level network extracts the motion trajectory features of the mechanical arm, and the third level network synthesizes the features of the previous two levels to generate a load state evaluation value;

[0024] According to the load state evaluation value, the weight coefficients of each feature channel are calculated through a nonlinear mapping function, wherein the weight of the pressure feature channel increases with the increase of the load, and the weight of the displacement feature channel decreases with the increase of the motion speed;

[0025] The weighted multi-channel features are input into a control quantity generation module, and a first control feature value is output through a preset load-control quantity conversion relationship table, and the conversion relationship table is periodically updated according to historical working condition data.

[0026] As a further scheme of the present application, the dynamic weight adjustment process further comprises the following optimization steps:

[0027] A load feature memory bank based on a sliding time window is established to store the feature signal change trajectories of the last N sampling periods;

[0028] A pattern matching algorithm is used to compare the similarity of the current feature signal with the typical working condition patterns in the memory bank, and the weight combination corresponding to the highest matching degree of the historical pattern is selected as the initial value;

[0029] The initial weight is fine-tuned online through an incremental learning method, and the adjustment amplitude is proportional to the degree of deviation of the current working condition from the typical pattern, and the finally generated first control feature value contains a static compensation component and a dynamic adjustment component.

[0030] As a further scheme of the present application, the differential order adjustable phase compensation process comprises the following specific steps:

[0031] A differential order adjustment mechanism based on impact intensity grading is established, the amplitude of the high-frequency feature signal is divided into three intensity intervals, and different differential orders are configured respectively: shallow order differential is used in the low amplitude interval to maintain system stability, medium order differential is used in the medium amplitude interval to achieve fast response, and deep order differential is used in the high amplitude interval for strong suppression;

[0032] The differential signal after order adjustment is subjected to phase calibration processing, and the phase lag introduced by the signal processing link is eliminated through a delay compensation unit;

[0033] The phase-corrected differential signal is multiplied with the original high-frequency characteristic signal to generate a second control characteristic value with pressure oscillation suppression effect, which contains both pressure change trend information and instantaneous fluctuation intensity information.

[0034] As a further scheme of the present application, the differential order adjustable phase compensation process further comprises the following optimization steps:

[0035] A pressure impact characteristic database is constructed to record the waveform patterns of high-frequency characteristic signals under different working conditions and the corresponding optimal differential order parameters;

[0036] Real-time pattern recognition technology is used to match the current high-frequency characteristic signal with the typical waveforms in the characteristic database, and the differential order combination is dynamically selected according to the matching result;

[0037] The generated second control characteristic value is evaluated for effect through a closed-loop verification mechanism, and when the pressure oscillation suppression effect does not meet the expectation, the differential order reconfiguration process is automatically triggered until a stable pressure suppression effect is obtained.

[0038] As a further scheme of the present application, the drive control model process comprises the following specific steps:

[0039] A dual-channel feature fusion module is constructed to set a main action channel for the first control characteristic value and an auxiliary correction channel for the second control characteristic value, the main channel uses fixed gain transmission, and the auxiliary channel uses variable gain transmission, and the gain coefficient is dynamically adjusted according to the current pressure fluctuation amplitude;

[0040] A command smoothing processor is cascaded after the feature fusion module to perform slope limiting on the fused control command to limit the command change rate to not exceed the dynamic response capability of the hydraulic actuator;

[0041] A actuator characteristic compensation unit is used to perform nonlinear correction on the generated adjustment command to compensate for the dead zone characteristics of the hydraulic valve and the coupling effect of flow and pressure, and output the final executable actuator drive command.

[0042] As a further scheme of the present application, the drive control model process further comprises the following optimization steps:

[0043] A mapping database of control characteristics and execution effects is established to store the corresponding relationship between historical control characteristic values and actual execution effects;

[0044] A feedforward-feedback composite control architecture is used, the feedforward branch directly generates a reference command according to the current control characteristic value, and the feedback branch performs fine tuning correction according to the execution effect deviation;

[0045] The setting instruction safety check link is arranged, when the generated adjusting instruction exceeds the working range of the actuator, the instruction reconstruction mechanism is started, and the characteristic fusion proportion is automatically adjusted under the premise of maintaining the control effect.

[0046] An adaptive control system for a hydraulic driving guniting manipulator, comprising:

[0047] A signal acquisition module is arranged for collecting pressure fluctuation signals, flow change signals and mechanical arm displacement signals of the hydraulic system in real time.

[0048] A frequency domain decomposition module is arranged for carrying out frequency domain decomposition processing on the collected signals, and separating out low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals reflecting pressure shocks.

[0049] A dynamic weight adjustment module is arranged for carrying out dynamic weight adjustment processing on the low-frequency characteristic signals, and generating first control characteristic values for compensating load changes.

[0050] A phase compensation processing module is arranged for carrying out phase compensation processing with adjustable differential order on the high-frequency characteristic signals, and generating second control characteristic values for suppressing pressure oscillation.

[0051] A drive control module is arranged for inputting the first control characteristic values and the second control characteristic values into a drive control model of the guniting manipulator, and generating adjusting instructions of the hydraulic actuator.

[0052] The beneficial effects of the present application are as follows:

[0053] (1) In the present application, through the integrated real-time signal acquisition system, the dynamic information of the pressure fluctuation, flow change and mechanical arm displacement of the hydraulic system is accurately obtained, and advanced frequency domain decomposition technology is used to subdivide these complex composite signals into low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals revealing pressure impact characteristics. For these two different signals, we have designed a dynamic weight adjustment strategy and a differential order adjustable phase compensation mechanism: the former realizes sensitive response and accurate compensation to load changes through intelligent weight distribution of low-frequency signals; the latter uses an innovative fractional differential processing method to effectively suppress transient pressure oscillation caused by rapid operation. This meticulous signal processing method not only greatly improves the adaptability and response speed of the system to external working condition changes, but also significantly enhances the overall industrial control precision, enabling the shotcrete manipulator to maintain efficient and stable operation performance in a variable working environment. In addition, combined with advanced signal processing algorithms such as adaptive threshold noise reduction and cross-correlation analysis, the accuracy and reliability of the data are further ensured, providing a solid foundation for more refined industrial control. This series of innovative measures collectively marks a major breakthrough in the control technology of hydraulic-driven shotcrete manipulators, opening up new paths for improving engineering operation efficiency and quality.

[0054] (2) The present application deeply integrates various advanced signal processing technologies and intelligent control algorithms, and constructs a highly adaptive, self-learning and self-optimizing hydraulic drive spraying manipulator control system. In the signal processing layer, the system adopts adaptive threshold denoising technology, combines wavelet packet transform and sliding window normalization processing, can dynamically identify and effectively suppress environmental noise and non-stationary interference, significantly improves the signal-to-noise ratio and the accuracy of feature extraction; At the same time, through the cross-correlation analysis module, the coupling relationship between low-frequency and high-frequency signals is monitored in real time, the abnormal coupling is identified in time, and the frequency domain selective attenuation is implemented, so as to guarantee the physical consistency of signal processing and system stability. In terms of intelligent control, the system introduces a three-level cascaded neural network architecture, combines dynamic weight adjustment mechanism and incremental learning algorithm, can adaptively optimize the control parameters according to the historical working mode and real-time load state, realize the closed-loop optimization from experience learning to online adjustment, greatly enhance the environmental adaptability and control robustness of the system. In addition, the system also integrates a closed-loop verification mechanism and an instruction safety verification link, the former realizes dynamic reconfiguration of parameters through continuous evaluation of control effect, and the latter ensures the safety and feasibility of the output instruction through multi-level verification strategy, effectively prevents misoperation and system out of control. In the module maintenance layer, the system has automatic calibration, real-time fault self-diagnosis and data memory update functions, not only can regularly calibrate the frequency response characteristics and phase consistency of the processing channel, but also can quickly locate the fault source and enable the redundant or estimation algorithm when an abnormality occurs, guarantee the reliability and stability of the system for long-time continuous operation. In summary, these integrated, intelligent and adaptive technical means and mechanisms work together, so that the present application can still maintain efficient, stable and safe operation state when facing complex and variable working environment, significantly improves the performance and intelligent level of the spraying manipulator in actual engineering application. BRIEF DESCRIPTION OF DRAWINGS

[0055] The present application will be further described below in conjunction with the drawings.

[0056] Figure 1 is a flow chart of an adaptive control method for a hydraulic drive spraying manipulator of the present application;

[0057] Figure 2 is a flow chart of an adaptive control system for a hydraulic drive spraying manipulator in the present application. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0059] Referring to Figure 1 As shown in the drawings, the present application is an adaptive control method for a hydraulic driving guniting manipulator, comprising the following steps:

[0060] S1: Real-time acquisition of pressure fluctuation signals, flow change signals and mechanical arm displacement signals of the hydraulic system;

[0061] S2: Frequency domain decomposition processing is performed on the collected signals to separate out low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals reflecting pressure shocks;

[0062] S3: Dynamic weight adjustment processing is performed on the low-frequency characteristic signals to generate first control characteristic values for compensating for load changes;

[0063] S4: The high-frequency characteristic signals are subjected to phase compensation processing with adjustable differential order to generate second control characteristic values for suppressing pressure oscillation;

[0064] S5: The first control characteristic values and the second control characteristic values are input into the driving control model of the guniting manipulator to generate adjustment instructions for the hydraulic actuator.

[0065] In S1, real-time acquisition of pressure fluctuation signals, flow change signals and mechanical arm displacement signals of the hydraulic system is performed, specifically including:

[0066] The signal acquisition module serves as the input end of the system and is mainly responsible for real-time acquisition of key parameters during the operation of the hydraulic system. The module includes three independent sensing units: a pressure sensing unit, a flow sensing unit and a displacement sensing unit, which are electrically isolated from each other to avoid signal crosstalk. The pressure sensing unit uses a diffused silicon pressure sensor installed at the inlet and outlet of the hydraulic cylinder, with a measurement range of 0-40 MPa and a sampling frequency set to 1 kHz, which can accurately capture the transient characteristics in the pressure fluctuation signal. The sensor output signal is preprocessed by a second-order Butterworth low-pass filter with a cutoff frequency of 500 Hz to eliminate high-frequency noise interference.

[0067] The flow sensing unit uses a turbine flowmeter installed in the main oil line with a measurement accuracy of ±0.5%FS. To adapt to the characteristics of rapid flow changes in guniting operations, the flow signal acquisition uses a dual sampling rate mechanism: a 100Hz base sampling rate is used under steady-state conditions, and when the flow change rate exceeds the set threshold, it automatically switches to a 1kHz high-speed sampling mode. The pulse signal output by the flowmeter is shaped by a Schmidt trigger and sent to a counter for processing, and finally converted into a standard analog signal.

[0068] The displacement sensing unit adopts a magnetostrictive linear displacement sensor to measure the real-time position of each joint of the robot arm. The sensor is installed inside the hydraulic cylinder piston rod, with a resolution of 0.01 mm and an absolute measurement error of no more than ±0.1 mm. To eliminate the influence of mechanical vibration on the measurement results, the displacement signal is subjected to moving average filtering processing, with the window length being adaptively adjusted according to the movement speed of the robot arm: a 20-point window is used to ensure measurement accuracy at low speed, and the window is reduced to 5 points at high speed to reduce phase lag.

[0069] The output signals of the three sensing units are transmitted to the main controller through the CAN bus with a fixed transmission period of 1 ms. To improve system reliability, the signal transmission adopts a double-redundancy design: the current value and the previous cycle value are sent simultaneously at each sampling time, and the historical data is automatically enabled when data anomalies are detected. The main controller receiving end is provided with a data checking mechanism to eliminate and interpolate abnormal values outside the reasonable range.

[0070] The signal acquisition module also includes an environmental parameter monitoring function, which monitors the hydraulic oil temperature in real time through a temperature sensor and automatically adjusts the signal processing parameters when the oil temperature exceeds the set threshold. For example, in high-temperature working conditions, the filter cutoff frequency of the pressure signal is appropriately increased to compensate for changes in sensor sensitivity. All collected raw data is marked with a timestamp to ensure strict synchronization of multi-channel signals.

[0071] To adapt to different working conditions, the signal acquisition module supports online parameter configuration. The operator can adjust the sampling frequency, filter parameters, and other settings of each channel through the human-machine interface, and the modified parameters take effect immediately and are stored in the non-volatile memory. The system automatically loads the last valid configuration when powered on, ensuring the continuity of parameter settings.

[0072] The signal acquisition module has a self-diagnosis function that periodically checks the working state of each sensor. When a sensor fault is detected, the system automatically switches to a backup sensor or enables an estimation algorithm to maintain system operation, while displaying alarm information on the operation interface. The diagnosis includes multiple indicators such as sensor power voltage, signal amplitude range, and change rate reasonableness to ensure the reliability of the collected data.

[0073] In S2, the collected signals are subjected to frequency domain decomposition processing to separate low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals reflecting pressure shocks, including:

[0074] The frequency domain decomposition module is the core processing unit of the application, mainly responsible for decomposing the collected composite signal into low-frequency characteristic signal and high-frequency characteristic signal with clear physical meaning. The module adopts a mixed signal processing architecture, combining traditional filtering technology and modern time-frequency analysis method, and can adapt to the complex and changeable working environment of the shotcrete robot. The module input end receives the original data stream from the signal acquisition module, first carries out signal quality detection, and marks and processes the data segment with missing or abnormal data.

[0075] The multi-stage filter bank is the first processing stage of frequency domain decomposition. The first stage Butterworth low-pass filter adopts a fourth-order design, and its cutoff frequency can be dynamically adjusted in the range of 50-150 Hz, with a default setting of 100 Hz. The parameters of the filter are configured digitally, and a zero-phase filter algorithm is used to avoid introducing time delay. The second stage comb filter bank is composed of eight parallel band-pass filters, with center frequencies set at key frequency points from 1 kHz to 5 kHz, and the bandwidth of each filter is set to 50 Hz, which can effectively extract the characteristic harmonic components of pressure impact. The output of the filter bank passes through a gain compensation circuit to ensure the amplitude consistency of each frequency band signal.

[0076] The sliding window normalization processing is implemented for the low-frequency characteristic signal. The processing unit adopts a ring buffer structure, and the window length is automatically adjusted according to the real-time motion speed of the robot arm: when the speed is below 0.1 m / s, a long window of 500 ms is used, when the speed is between 0.1-0.5 m / s, a medium window of 200 ms is used, and when the speed is higher than 0.5 m / s, a short window of 50 ms is used. The normalization processing includes two steps of direct current component removal and amplitude standardization. The processed signal keeps the original phase information unchanged, and the amplitude is uniformly adjusted to the standard working range.

[0077] The time domain envelope detector is responsible for the amplitude reconstruction of the high-frequency characteristic signal. The circuit adopts a scheme of precise full-wave rectification combined with a second-order low-pass filter, with a cutoff frequency of 200 Hz, which can completely retain the envelope characteristics of the impact waveform. To eliminate the carrier residual, an adaptive notch filter is added to the rear stage of the circuit, with the center frequency automatically tracking the main frequency component of the input signal. The reconstructed envelope signal passes through a sample-and-hold circuit to maintain a proper time at the impact peak, facilitating the capture of transient features by the subsequent processing module.

[0078] The wavelet packet transform unit operates as an auxiliary processing channel. The unit uses DB4 wavelet basis function to decompose the input signal into 16 layers, generating 16 frequency band detail coefficients. The low-frequency reconstruction selects the 3rd to 5th layer coefficients, and the high-frequency reconstruction selects the 10th to 12th layer coefficients, corresponding to the frequency range of 1 kHz to 4 kHz. The wavelet processing adopts an overlapping segmentation method, with each segment of data being 1024 points long and overlapping by 256 points, ensuring time domain continuity.

[0079] The adaptive threshold denoising process is implemented for the high-frequency signal of wavelet reconstruction. The algorithm estimates the noise level in real time and dynamically adjusts the threshold curve according to the signal power spectrum characteristics. In the steady state, the hard threshold processing is adopted to retain the significant impact components; in the transient state, the soft threshold mode is switched to smooth the small fluctuations. The threshold update period is 100 ms, which ensures timely tracking of the working condition changes. The denoised signal is phase corrected to compensate for the group delay introduced by wavelet transform.

[0080] The cross-correlation analysis module monitors the coupling relationship between low-frequency and high-frequency signals. The module calculates the normalized cross-correlation coefficient of the two signals, with a time window length of 1 second and a sliding step of 100 ms. When the correlation coefficient exceeds the threshold of 0.7, it is determined that there is abnormal coupling, and the compensation mechanism is started. The compensation process first analyzes the abnormal frequency band, then performs selective attenuation in the frequency domain, and finally reconstructs the time domain signal. The compensation parameters are stored in the non-volatile memory, forming a continuously optimized knowledge base.

[0081] The signal output interface packs and transmits the processed low-frequency feature signal and high-frequency feature signal. Each data packet contains a time series of 100 sampling points, accompanied by a time stamp and a signal quality identifier. The transmission protocol supports error retransmission mechanism to ensure data integrity. The output interface also provides real-time monitoring signals for external devices to verify the processing effect.

[0082] The module has a built-in self-calibration function, which periodically performs the following calibration procedures: first, inject a standard test signal to verify the frequency response characteristics of each processing channel; then perform white noise test to evaluate the noise suppression capability of the system; finally, implement step response test to check the transient processing performance. The calibration results automatically generate a report, and the abnormal situation triggers an alarm signal. The calibration period can be configured, and the default setting is to perform it once every 24 hours.

[0083] In S3, the low-frequency feature signal is processed by dynamic weight adjustment to generate a first control feature value for compensating for load changes, which specifically includes:

[0084] The dynamic weight adjustment module is one of the core processing units of the present application, which is specially responsible for intelligent weight distribution and feature fusion of the low-frequency feature signal. The module adopts a three-level cascaded neural network architecture, and each level of network has a specific feature extraction function. The first level network is configured as a double-input single-output structure, which is specially used to process the pressure-flow coupling relationship of the hydraulic system. The network contains 32 processing nodes and uses Sigmoid activation function, which can accurately capture the nonlinear dynamic characteristics between pressure and flow. The preprocessing unit at the network input end will perform standardization processing on the original signal to eliminate the influence of dimension difference.

[0085] The second level network focuses on the extraction of the mechanical arm motion trajectory features. The network adopts a time series processing structure, with an input window length of 50 sampling points, and can identify dynamic features such as the speed and acceleration of the mechanical arm motion. The network contains long short-term memory units, which can remember typical motion patterns. During the feature extraction process, the network automatically ignores minor fluctuations caused by measurement noise and focuses on identifying the true motion intention. The network output end is connected to a feature selector, which only retains the most representative motion features.

[0086] The third level network serves as a feature fusion center, intelligently combining the features extracted by the previous two networks. The network adopts a fully connected structure, containing three processing layers with node numbers of 64, 32, and 16 respectively. The dropout technique is used during network training to prevent overfitting, with a retention probability of 0.7. The fused features are converted into load state evaluation values through normalization processing. This value is a dimensionless parameter, with a range limited to 0 to 1, and a larger value indicates a heavier load.

[0087] The weight coefficient calculation unit dynamically adjusts the importance weight of each feature channel based on the load state evaluation value. The unit has a built-in nonlinear mapping function, which is implemented through a lookup table for fast calculation. The weight calculation of the pressure feature channel uses an increasing curve, which grows slowly in the light load area and rapidly in the heavy load area. The weight calculation of the displacement feature channel uses a decreasing curve, which maintains a high weight at low speeds and gradually decreases at high speeds. The weight adjustment process introduces an inertia element to avoid control oscillation caused by sudden weight changes.

[0088] The core of the control quantity generation module is the load-control quantity conversion table. This table uses a three-dimensional structure, with the load state evaluation value, motion speed, and pressure change rate as the index dimensions. The table data is maintained through a combination of offline training and online learning, with initial values derived from historical optimal parameters and continuously optimized based on actual control effects during operation. The table query uses a trilinear interpolation algorithm to ensure smooth and continuous output. At the end of each working period, the module automatically records the optimal control parameters for incremental updating of the table.

[0089] The load feature memory bank uses a ring buffer structure with a storage capacity of 1000 working periods. Each storage unit contains a complete feature vector, weight combination, and control effect evaluation. Data storage automatically performs compression processing to preserve key features while saving storage space. The memory bank maintenance process periodically performs data organization, deletes records with excessively high repetition, and retains the most representative working condition patterns.

[0090] The mode matching algorithm adopts an improved dynamic time warping technique, which can effectively process work segments of different lengths. The matching process considers the time domain shape and statistical characteristics of the feature vector, and calculates a comprehensive similarity score. The algorithm sets a matching quality threshold, and when the score of the optimal match is below the threshold, it automatically switches to a default weight combination to ensure system reliability. The matching result is accompanied by a confidence index for the subsequent processing module to reference.

[0091] The incremental learning mechanism adopts a small batch update strategy, with an adjustment amplitude not exceeding 10% of the original value. The learning process is subject to double constraints: both quickly adapting to working condition changes and maintaining system stability. The direction of weight fine-tuning is determined by the trend of control effectiveness in the last 10 cycles, and the moving average method is used to eliminate random interference. The learning algorithm has a forgetting factor built in, which automatically reduces the influence weight of old data to ensure that the system always tracks the current optimal working point.

[0092] The first control characteristic value generated by the module output interface consists of two components: the static compensation component is derived from the reference value of the conversion relationship table, and the dynamic adjustment component reflects the real-time learning result. The mixing ratio of the two components is automatically adjusted according to the stability of the working condition, with a focus on the static component in steady state and an increased dynamic component proportion in transient state. The output data is accompanied by a quality flag bit to identify the confidence level of the current control parameters.

[0093] In S4, a phase compensation processing with adjustable differential order is performed on the high-frequency characteristic signal to generate a second control characteristic value for suppressing pressure oscillation, specifically including:

[0094] The phase compensation processing module is a key technical module of the present application, which is responsible for processing high-frequency characteristic signals and generating control characteristic values for suppressing pressure oscillation. The module uses an innovative adjustable differential order technology, which can intelligently adjust the processing strategy according to the impact strength. The module input receives high-frequency characteristic signals from the frequency domain decomposition module, first performs signal integrity check to ensure data validity. The input buffer uses a double buffer design to realize uninterrupted data stream processing.

[0095] The impact strength grading unit is one of the core components of this module. This unit monitors the amplitude characteristics of the input signal in real time and divides it into three working intervals: the low amplitude interval corresponds to 0-20% of the range, the medium amplitude interval corresponds to 20-60% of the range, and the high amplitude interval corresponds to 60-100% of the range. Each interval is configured with independent differential order parameters, with a shallow order of 0.3-0.5 for the low interval, a medium order of 0.5-0.7 for the medium interval, and a deep order of 0.7-0.9 for the high interval. The interval switching sets a 5% hysteresis band to prevent boundary oscillation. The grading result is updated in real time by the state register for subsequent processing units to call.

[0096] The differential processor implements differential operation with adjustable order. The processor uses digital filtering to approximate fractional order differentiation, and the core algorithm is based on the principle of weighted moving average. During processing, the differential order can be dynamically adjusted according to the classification results of shock intensity, and the adjustment process is smooth to avoid step changes. The processor has an anti-aliasing filter built-in, and the cutoff frequency is automatically matched to the current differential order to ensure signal quality. The output of the differential operation is amplitude normalized to eliminate the gain fluctuations caused by order changes.

[0097] The delay compensation unit is responsible for correcting the phase deviation of the signal processing link. The unit uses an all-pass filter structure, and the phase compensation amount can be programmed. The compensation parameters are obtained through offline calibration and stored in a calibration table. During real-time processing, the corresponding compensation parameters are automatically found according to the current differential order. The phase error of the compensated signal is controlled within ±5 degrees, meeting the accuracy requirements of the control system. The unit has a self-checking function that verifies the phase characteristics periodically.

[0098] The signal coupler intelligently fuses the processed differential signal with the original high-frequency features. The fusion algorithm uses conditional weighted multiplication, focusing on trend information when the signal amplitude is small and strengthening fluctuation components when the amplitude is large. The coupling coefficient is dynamically adjusted according to the shock intensity to ensure that the output characteristic value has clear physical meaning. The coupling process preserves the polarity characteristics of the original signal to avoid information distortion. The output end is configured with a limiter to constrain the characteristic value within a safe range.

[0099] The pressure shock feature database is the basis for intelligent learning of this module. The database uses a hierarchical storage structure, with the upper layer storing typical waveform features and the lower layer recording corresponding optimal processing parameters. During data collection, the system automatically selects representative work segments and extracts time-domain and frequency-domain feature vectors. Principal component analysis is used for feature extraction to reduce data dimensionality. The database capacity is designed to be 1000 records, and the least recently used strategy is used for space management.

[0100] The real-time pattern recognition engine uses an improved dynamic time warping algorithm. The engine performs pattern matching every 50ms, calculating the similarity between the current signal and the database templates. The matching process considers multiple indicators such as waveform shape, spectral distribution, and statistical features. To improve real-time performance, the engine uses a hierarchical search strategy, first performing coarse screening and then fine matching. The recognition result is accompanied by a confidence score, and only matches with a score higher than the threshold are adopted.

[0101] The closed-loop verification mechanism constitutes the quality assurance system of the module. The mechanism continuously monitors the suppression effect of pressure oscillation, and the evaluation indexes include peak decay rate, stabilization time, and overshoot, etc. The verification period is 100 ms, and a sliding window statistics is adopted. When the effect is not up to the standard, the parameter reconfiguration process is triggered. The reconfiguration process adopts gradient trial method, and gradually adjusts the differential order until a satisfactory effect is obtained. The result of each reconfiguration is fed back to update the feature database, and continuous optimization is realized.

[0102] The module output interface generates a second control feature value, which is a signed number, reflecting both the direction of pressure change and the intensity of oscillation. The output data is accompanied by a quality mark indicating the confidence level of the current processing. The interface protocol supports real-time monitoring function, allowing external devices to read intermediate processing results. The working state of the module is intuitively displayed through indicator lights, including normal operation, learning update, and abnormal alarm.

[0103] In S5, the first control feature value and the second control feature value are input into a driving control model of the gunning manipulator to generate a regulation instruction of a hydraulic actuator, specifically including:

[0104] The driving control module, as the final output link of the application, is responsible for converting the processed feature signal into an executable hydraulic control instruction. The module adopts a dual-channel hybrid control architecture, with the main channel processing steady-state load compensation requirements and the auxiliary channel coping with dynamic pressure oscillation. During initialization, the module first loads the default parameter configuration, including key parameters such as basic gain coefficient and amplitude threshold. These parameters are stored in non-volatile memory to ensure that power failure does not result in loss.

[0105] The dual-channel feature fusion module realizes signal synthesis using a weighted summation algorithm. The main channel is configured with a fixed gain amplifier, with the gain coefficient set to an optimal value within the range of 0.7-0.9 to ensure the dominance of the basic control quantity. The auxiliary channel is configured with an automatic gain regulator, whose gain coefficient is dynamically calculated based on the real-time pressure fluctuation rate. The larger the fluctuation rate, the higher the gain, with a maximum of 0.5. Both signals are subjected to low-pass filtering before synthesis, with the main channel cutoff frequency set to 10 Hz and the auxiliary channel set to 50 Hz, ensuring dynamic response and avoiding high-frequency interference. The synthesis operation uses a saturation adder to prevent output overflow.

[0106] The instruction smoothing processor adopts a digital slope limiting algorithm. The processor calculates the gradient of the control instruction in real time, and automatically inserts a transition instruction point when the instantaneous change rate exceeds the set threshold. The threshold parameter is set according to the dynamic characteristics of the actuator, and for a typical hydraulic servo valve, the limit is 5-10 times per second at full range. The processor has a built-in prediction function, which estimates the next time requirement based on the instruction trend of the previous three periods, and performs smoothing processing in advance. In special cases, when an emergency control requirement is detected, the restriction condition can be temporarily relaxed to prioritize system safety.

[0107] The actuator characteristic compensation unit contains two main functional blocks: dead zone compensator and coupling demodulator. The dead zone compensator uses a lookup table method to predict the actual opening of the valve core based on the current instruction value and historical motion direction, and the compensation amount is automatically adjusted over time to adapt to changes in valve wear. The coupling demodulator monitors the pressure-flow relationship of the system and injects a decoupling compensation signal when a strong coupling condition is identified. The compensation parameters are updated regularly through an online identification algorithm to ensure adaptation to changes in characteristics under different oil temperatures and working pressures.

[0108] The control feature-actuation effect mapping database uses a time series data structure for storage. Each record contains control feature vector, output instruction, and actual actuation effect information, with a time resolution of 10ms. The database has an automatic sorting mechanism that regularly merges similar records and deletes redundant data. The query interface supports multi-dimensional indexing, allowing for fast retrieval of historical optimal control strategies. The database capacity is designed to store the last 100 hours of working data, using a first-in, first-out management strategy.

[0109] The feedforward-feedback composite control architecture achieves precise regulation. The feedforward controller directly analyzes the current feature vector and generates a reference instruction through fuzzy reasoning, with a response time of less than 5ms. The feedback regulator uses an incremental PID algorithm to fine-tune the control amount based on the actuation deviation, with a regulation period of 20ms. The outputs of the two branches are weighted and fused in the synthesizer, with the weight coefficients adjusted dynamically according to the working conditions. In steady state, the feedforward is emphasized, while in transient state, the feedback is strengthened. Anti-saturation protection is provided at the output end of the synthesizer to avoid integral saturation.

[0110] The instruction safety verification link implements three levels of protection. The first level checks the numerical range to ensure that the instruction is within the hardware allowed range; the second level verifies the reasonableness of the change trend to prevent abnormal jumps; and the third level evaluates the execution prediction effect to avoid dangerous operations. When any level of check fails, the instruction reconstruction process is immediately started. The reconstruction algorithm preserves the control intent of the original instruction and generates a safe and feasible alternative instruction through methods such as equal proportion scaling or piecewise approximation. The reconstruction process is recorded in real time for fault analysis.

[0111] The module output interface adopts an industry standard protocol, supports 4-20 mA analog output and CAN bus digital output in two modes. The output driver has an overcurrent protection function to ensure the safety of the actuator. The interface circuit adopts an optical isolation design to improve the anti-interference ability. After each output cycle ends, the module automatically generates a running log to record key parameters and abnormal events, and the log data can be read through the debugging interface.

[0112] The module maintenance function includes automatic calibration and fault self-diagnosis. The automatic calibration program is executed periodically to verify the performance indicators of each processing link by injecting test signals. The fault diagnosis system monitors the module working state in real time, and the detection items include signal continuity, processing timeliness and logic rationality, etc. The diagnosis results are divided into three levels of attention, warning and serious, corresponding to different processing strategies. All maintenance operations are recorded in detail logs to support post-analysis.

[0113] Referring to Figure 2 An adaptive control system for a hydraulic-driven guniting manipulator, comprising:

[0114] A signal acquisition module for acquiring pressure fluctuation signals, flow change signals and mechanical arm displacement signals of the hydraulic system in real time;

[0115] A frequency domain decomposition module for frequency domain decomposition processing of the acquired signals, separating out low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals reflecting pressure shocks;

[0116] A dynamic weight adjustment module for dynamic weight adjustment processing of the low-frequency characteristic signals to generate first control characteristic values for compensating load changes;

[0117] A phase compensation processing module for differential order adjustable phase compensation processing of the high-frequency characteristic signals to generate second control characteristic values for suppressing pressure oscillation;

[0118] A drive control module for inputting the first control characteristic values and the second control characteristic values into a drive control model of the guniting manipulator to generate adjustment instructions of the hydraulic actuator.

[0119] Working principle of the present application: The present application realizes precise control under complex working conditions through multi-modal signal fusion and intelligent control technology, including: first, real-time acquisition of pressure, flow and mechanical arm displacement signals of the hydraulic system through high-precision sensors, using multi-stage filtering and wavelet packet transform combined frequency domain decomposition technology to separate the original signal into low-frequency characteristic signal reflecting load change and high-frequency characteristic signal representing pressure impact; For low-frequency signals, a three-level neural network feature extraction architecture is constructed, and load compensation control is generated through dynamic weight adjustment, wherein the weight coefficient is nonlinearly mapped according to the load state evaluation value, and online optimization is combined with historical working condition mode matching; The high-frequency signal is subjected to a differential order adjustable processing based on impact intensity grading, and an oscillation suppression control quantity is generated through phase compensation and signal coupling, and the differential order is dynamically adjusted according to the real-time mode recognition result; Finally, a double-channel feature fusion strategy is used to intelligently integrate the two types of control quantities, and the hydraulic regulation command is output after instruction smoothing processing and actuator characteristic compensation. The system innovatively introduces a control feature-implementation effect mapping database and a feedforward-feedback composite control architecture, and continuously optimizes the control performance through closed-loop verification and parameter self-learning mechanism. This method effectively solves the technical problem that load adaptability and oscillation suppression cannot be considered in traditional control, and significantly improves the stability and control precision of the spraying operation.

[0120] The above describes one embodiment of the present application in detail, but the content described is only the preferred embodiment of the present application, and cannot be considered as limiting the scope of the present application. Any equivalent changes and improvements made within the scope of the present application should still be within the scope of the present application.

Claims

1. An adaptive control method for a hydraulic drive guniting robot, characterized in that, The method comprises the following steps: S1: Real-time acquisition of pressure fluctuation signal, flow change signal and mechanical arm displacement signal of the hydraulic system; S2: Frequency domain decomposition processing is performed on the collected signals to separate out low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals reflecting pressure shocks; S3: Dynamic weight adjustment processing is performed on the low-frequency characteristic signals to generate a first control characteristic value for compensating for load changes; The dynamic weight adjustment processing comprises the following specific steps: A feature extraction network with three levels of hidden layers is constructed, the first level of network extracts pressure-flow coupling characteristics of the hydraulic system, the second level of network extracts mechanical arm motion trajectory characteristics, and the third level of network comprehensively generates a load state evaluation value based on the previous two levels of characteristics; According to the load state evaluation value, the weight coefficients of each characteristic channel are calculated through a nonlinear mapping function, wherein the weight of the pressure-flow coupling characteristic channel increases with the increase of the load, and the weight of the mechanical arm motion trajectory characteristic channel decreases with the increase of the motion speed; The weighted pressure-flow coupling characteristics and mechanical arm motion trajectory characteristics are input into a control quantity generation module, and a first control characteristic value is output through a preset load-control quantity conversion relationship table; The dynamic weight adjustment processing further comprises the following optimization steps: A load characteristic memory bank based on a sliding time window is established to store the low-frequency characteristic signal change trajectories of the last N sampling periods; A pattern matching algorithm is used to compare the similarity of the low-frequency characteristic signals with the typical working condition patterns in the characteristic memory bank, and the combination of the weights of the pressure-flow coupling characteristic channel and the mechanical arm motion trajectory characteristic channel corresponding to the typical working condition pattern with the highest matching degree is selected as the initial weight; The initial weight is fine-tuned online through an incremental learning method, and the adjustment amplitude is proportional to the degree of deviation of the current working condition from the typical working condition pattern, and the finally generated first control characteristic value contains a static compensation component and a dynamic adjustment component; S4: The high-frequency characteristic signals are subjected to phase compensation processing with adjustable differential order to generate a second control characteristic value for suppressing pressure oscillation; S5: The first control characteristic value and the second control characteristic value are input into the drive control model of the shotcrete mechanical arm to generate the adjustment instruction of the hydraulic actuator.

2. The self-adaptive control method for a hydraulic drive guniting manipulator according to claim 1, characterized in that, The frequency domain decomposition processing comprises the following specific steps: A signal separation device based on multi-stage filters is constructed, a first stage adopts a Butterworth filter with adjustable cutoff frequency to extract 0-100Hz band signals as low-frequency characteristic signals, and a second stage adopts a comb filter group to extract harmonic components of specific frequency points greater than 1kHz as high-frequency characteristic signals; The low-frequency characteristic signals are subjected to sliding window normalization processing, and the window length is dynamically adjusted according to the mechanical arm motion speed; The high-frequency characteristic signals are subjected to amplitude reconstruction based on time domain envelope detection.

3. The self-adaptive control method for a hydraulic drive guniting manipulator according to claim 1, wherein, The frequency domain decomposition processing further comprises the following optimization steps: Wavelet packet transform is used to decompose the original signals for 16 layers, the 3rd-5th layer detail coefficients are selected to reconstruct the low-frequency characteristic signals, and the 10th-12th layer detail coefficients are selected to reconstruct the high-frequency characteristic signals; The reconstructed high-frequency characteristic signals are subjected to adaptive threshold denoising processing, and the threshold size is dynamically adjusted according to the signal-to-noise ratio under the current working condition; The cross-correlation function of the low-frequency characteristic signal and the high-frequency characteristic signal is established, and when the correlation coefficient exceeds a set threshold, a signal compensation mechanism is started to correct the low-frequency characteristic signal and the high-frequency characteristic signal.

4. The self-adaptive control method for a hydraulic drive guniting manipulator according to claim 1, wherein, The differential order adjustable phase compensation processing comprises the following specific steps: A differential order adjustment mechanism based on pressure impact intensity grading is established, the amplitude of the high-frequency characteristic signal is divided into three intensity intervals, and different differential orders are configured respectively: low-amplitude interval adopts shallow order differential to maintain system stability, medium-amplitude interval adopts medium order differential to realize fast response, and high-amplitude interval adopts deep order differential for strong suppression; The differential signal after order adjustment is subjected to phase calibration processing; The differential signal after phase calibration is coupled with the original high-frequency characteristic signal to generate a second control characteristic value.

5. The self-adaptive control method for a hydraulic drive guniting manipulator according to claim 1, wherein, The differential order adjustable phase compensation processing further comprises the following optimization steps: A pressure impact characteristic database is constructed to record the waveform mode of the high-frequency characteristic signal under different working conditions and the corresponding optimal differential order parameters; Real-time pattern recognition technology is used to match the current high-frequency characteristic signal with the typical waveforms in the pressure impact characteristic database, and the differential order combination is dynamically selected according to the matching result; Through a closed-loop verification mechanism, the generated second control characteristic value is evaluated for effect, and when the pressure oscillation suppression effect does not meet the expectation, the differential order reconfiguration process is automatically triggered until a stable pressure oscillation suppression effect is obtained.

6. The self-adaptive control method for a hydraulic drive guniting manipulator according to claim 1, wherein, The drive control model processing comprises the following specific steps: A double-channel feature fusion module is constructed, a main action channel of the first control characteristic value and an auxiliary correction channel of the second control characteristic value are set, the main channel adopts fixed gain transmission, the auxiliary channel adopts variable gain transmission, and the gain coefficient is dynamically adjusted according to the current pressure fluctuation amplitude; A command smoothing processor is cascaded after the feature fusion module to perform slope amplitude limiting processing on the fused adjustment command, so that the change rate of the adjustment command does not exceed the dynamic response capability of the hydraulic actuator; Through an actuator characteristic compensation unit, the generated adjustment command is subjected to nonlinear correction to compensate for the dead zone characteristics of the hydraulic valve and the coupling effect of flow and pressure, and the final executable adjustment command is output.

7. The self-adaptive control method for a hydraulic drive guniting manipulator according to claim 1, wherein, The drive control model processing further comprises the following optimization steps: A mapping database of the first control characteristic value, the second control characteristic value and the execution effect is established to store the corresponding relationship between the historical first control characteristic value, the historical second control characteristic value and the actual execution effect; A feedforward-feedback composite control architecture is adopted, the feedforward branch directly generates a reference command according to the current first control characteristic value and the current second control characteristic value, and the feedback branch performs fine tuning correction according to the execution effect deviation; A command safety verification link is set, when the generated adjustment command exceeds the working range of the hydraulic actuator, a command reconstruction mechanism is started to automatically adjust the feature fusion ratio under the premise of maintaining the execution effect.

8. An adaptive control system for a hydraulically driven guniting robot, characterized by The adaptive control method for the hydraulic drive guniting manipulator according to any one of claims 1-7 comprises: A signal acquisition module, the signal acquisition module is used for real-time acquisition of pressure fluctuation signals, flow change signals and mechanical arm displacement signals of a hydraulic system; A frequency domain decomposition module is configured to perform frequency domain decomposition on the collected signals to separate a low-frequency characteristic signal reflecting load changes and a high-frequency characteristic signal reflecting pressure shocks; A dynamic weight adjustment module is configured to perform dynamic weight adjustment on the low-frequency characteristic signal to generate a first control characteristic value for compensating load changes; A phase compensation processing module is configured to perform phase compensation processing with adjustable differential order on the high-frequency characteristic signal to generate a second control characteristic value for suppressing pressure oscillation; A drive control module inputs the first control characteristic value and the second control characteristic value into a drive control model of the shotcrete robot to generate an adjustment instruction of the hydraulic actuator.

Images