Self-adaptive control method and system for hydraulically-driven guniting manipulator

Abstract

The invention relates to the technical field of hydraulically-driven industrial control, and particularly discloses a self-adaptive control method and system for a hydraulically-driven guniting manipulator, which are characterized in that a pressure fluctuation signal, a flow change signal and a mechanical arm displacement signal of a hydraulic system are collected in real time, and a frequency domain decomposition processing technology is combined to realize self-adaptive control of the hydraulically-driven guniting manipulator. Separating the composite signal into a low-frequency characteristic signal reflecting load change and a high-frequency characteristic signal reflecting pressure impact; for a low-frequency signal, a dynamic weight adjustment strategy is adopted to generate a load compensation control quantity; for high-frequency signals, pressure oscillation is suppressed through differential order adjustable phase compensation processing; and finally, inputting the two paths of control characteristic values into a driving control model of the guniting manipulator, and generating an adjusting instruction of the hydraulic actuating mechanism.

Description

Technical Field

[0001] The present invention relates to the technical field of hydraulically driven industrial control, and in particular to an adaptive control method and system for a hydraulically driven spraying manipulator. Background Art

[0002] In the field of modern industrial control automation, hydraulically driven manipulators are widely used in heavy-load, high-precision, or complex trajectory control operations, such as in harsh working environments such as construction spraying, mine support, and metallurgical casting. Traditional hydraulic control systems mostly use PID control or feedforward compensation strategies, relying on fixed parameter models, and are difficult to adapt to complex dynamic characteristics such as sudden load changes, external disturbances, and system nonlinearity. Especially in spraying operations, the frequent start and stop of the hydraulic system and the drastic fluctuations in flow and pressure can easily lead to problems such as system oscillation, response lag, and reduced control accuracy. In addition, the manipulator is affected by inertia, friction, and external load coupling during movement, making it difficult for industrial control models to accurately describe its dynamic behavior.

[0003] The existing technology has the following deficiencies:

[0004] This invention aims to address the technical challenge of traditional industrial control methods for hydraulically driven shotcrete manipulators, which face difficulties in achieving coordinated response and adaptive compensation of multi-band characteristic signals under the coupled effects of complex dynamic loads and high-frequency pressure shocks. Existing technologies typically treat pressure, flow, and displacement signals uniformly, ignoring the physical significance of signals in different frequency bands and their differential impact on system control. This results in control strategies experiencing response lag, overcompensation, or undercompensation when faced with sudden load changes and shock disturbances.

[0005] This invention uses frequency-domain decomposition technology to separate composite signals into low-frequency signatures reflecting load changes and high-frequency signatures reflecting pressure shocks. Dynamic weight adjustment and phase compensation with adjustable differential order are then employed to achieve differentiated processing and integrated control of multi-band signatures. This approach not only improves the system's responsiveness to complex operating conditions and industrial control accuracy, but also effectively suppresses pressure oscillations in hydraulic systems, filling a technological gap in the current industrial control field for the coordinated control of multi-band signatures. Summary of the Invention

[0006] The object of the present invention is to provide an adaptive control method and system for a hydraulically driven shotcrete manipulator to solve the above-mentioned problems.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] An adaptive control method for a hydraulically driven shotcrete manipulator comprises the following steps:

[0009] S1: Real-time acquisition of pressure fluctuation signals, flow change signals and robotic arm displacement signals of the hydraulic system;

[0010] S2: Perform frequency domain decomposition on the collected signal to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock;

[0011] S3: Performing dynamic weight adjustment processing on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes;

[0012] S4: performing phase compensation processing with adjustable differential order on the high-frequency characteristic signal to generate a second control characteristic value for suppressing pressure oscillation;

[0013] S5: Inputting the first control characteristic value and the second control characteristic value into the drive control model of the shotcrete manipulator to generate an adjustment instruction for the hydraulic actuator.

[0014] As a further solution of the present invention: the frequency domain decomposition processing includes the following specific steps:

[0015] A signal separation device based on a multi-stage filter was constructed. The first stage used a Butterworth filter with an adjustable cutoff frequency to extract the 0-100 Hz frequency band signal as the low-frequency characteristic signal. The second stage used a comb filter bank to extract the harmonic components of a specific frequency greater than 1 kHz as the high-frequency characteristic signal.

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

[0017] Amplitude reconstruction based on time domain envelope detection is performed on the high-frequency characteristic signal to retain the time-varying characteristics of the impulse waveform and eliminate the interference of the high-frequency carrier component.

[0018] As a further solution of the present invention: the frequency domain decomposition process further includes the following optimization steps:

[0019] The original signal is decomposed into 16 layers using wavelet packet transform, and the detail coefficients of the 3rd to 5th layers are selected to reconstruct the low-frequency feature signal, and the detail coefficients of the 10th to 12th layers are selected to reconstruct the high-frequency feature signal;

[0020] Adaptive threshold noise reduction is performed on the reconstructed high-frequency characteristic signal, and the threshold value is dynamically adjusted according to the signal-to-noise ratio under the current working conditions;

[0021] A cross-correlation function between the low-frequency characteristic signal and the high-frequency characteristic signal is established. When the correlation coefficient exceeds the set threshold, the signal compensation mechanism is activated to correct the decomposition result.

[0022] As a further solution of the present invention: the dynamic weight adjustment process includes the following specific steps:

[0023] A feature extraction network with three hidden layers was constructed. The first-level network extracted the pressure-flow coupling characteristics of the hydraulic system, the second-level network extracted the motion trajectory characteristics of the robotic arm, and the third-level network integrated the first two levels of features to generate the load state evaluation value.

[0024] According to the load state evaluation value, a weight coefficient of each characteristic channel is calculated by a nonlinear mapping function, wherein the weight of the pressure characteristic channel increases as the load increases, and the weight of the displacement characteristic channel decreases as the movement speed increases;

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

[0026] As a further solution of the present invention: the dynamic weight adjustment process further includes the following optimization steps:

[0027] Establish a load feature memory library based on a sliding time window to store the characteristic signal change trajectory of the latest N sampling cycles;

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

[0029] The initial weights are fine-tuned online through an incremental learning method, with the adjustment amplitude being proportional to the degree to which the current operating condition deviates from the typical mode. The first control characteristic value finally generated includes a static compensation component and a dynamic adjustment component.

[0030] As a further solution of the present invention: the differential order adjustable phase compensation process includes the following specific steps:

[0031] A differential order adjustment mechanism based on shock intensity grading is established, dividing the amplitude of the high-frequency characteristic signal into three intensity ranges, corresponding to different differential order configurations: shallow order differentials are used in the low amplitude range to maintain system stability, medium order differentials are used in the medium amplitude range to achieve fast response, and deep order differentials are used in the high amplitude range for strong suppression.

[0032] Perform phase calibration on the differential signal after order adjustment, and eliminate the phase lag introduced by the signal processing link through the delay compensation unit;

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

[0034] As a further solution of the present invention: the differential order adjustable phase compensation process further includes the following optimization steps:

[0035] Construct a pressure shock feature database to record the waveform patterns of high-frequency characteristic signals under different working conditions and their corresponding optimal differential order parameters;

[0036] Use real-time pattern recognition technology to match the current high-frequency characteristic signal with the typical waveform in the characteristic database, and dynamically select the differential order combination based on the matching results;

[0037] The generated second control eigenvalue is evaluated through a closed-loop verification mechanism. When the pressure oscillation suppression effect does not meet expectations, the differential order reconfiguration process is automatically triggered until a stable pressure suppression effect is achieved.

[0038] As a further solution of the present invention: the drive control model processing includes the following specific steps:

[0039] Construct a dual-channel feature fusion module, set the main action channel of the first control eigenvalue and the auxiliary correction channel of the second control eigenvalue, 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;

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

[0041] The generated adjustment instruction is corrected nonlinearly by the actuator characteristic compensation unit to compensate for the dead zone characteristics of the hydraulic valve and the coupling effect of flow and pressure, and the final executable mechanism drive instruction is output.

[0042] As a further solution of the present invention: the drive control model processing further includes the following optimization steps:

[0043] Establish a mapping database between control characteristics and execution effects, and store the corresponding relationship between historical control characteristic values ​​and actual execution effects;

[0044] Adopting a feedforward-feedback composite control architecture, the feedforward branch directly generates a benchmark instruction based on the current control characteristic value, and the feedback branch makes fine adjustments based on the execution effect deviation;

[0045] An instruction safety verification link is set up. When the generated adjustment instruction exceeds the working range of the actuator, the instruction reconstruction mechanism is activated to automatically adjust the feature fusion ratio while maintaining the control effect.

[0046] An adaptive control system for a hydraulically driven shotcrete manipulator, comprising:

[0047] A signal acquisition module, which is used to collect 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 used to perform frequency domain decomposition processing on the collected signal to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock;

[0049] a dynamic weight adjustment module, configured to perform dynamic weight adjustment processing on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes;

[0050] a phase compensation processing module, the phase compensation processing module being used 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;

[0051] A drive control module inputs the first control characteristic value and the second control characteristic value into a drive control model of the shotcrete manipulator to generate an adjustment instruction for the hydraulic actuator.

[0052] Beneficial effects of the present invention:

[0053] (1) In this invention, an integrated real-time signal acquisition system is used to accurately obtain the dynamic information of the hydraulic system's pressure fluctuations, flow changes, and manipulator displacement. Advanced frequency domain decomposition technology is then used to subdivide these complex composite signals into low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals revealing pressure shock characteristics. For these two signals of different nature, we have designed a dynamic weight adjustment strategy and a phase compensation mechanism with adjustable differential order. The former achieves sensitive response and precise compensation to load changes by intelligently allocating weights to low-frequency signals. The latter effectively suppresses transient pressure oscillations caused by rapid operation using an innovative fractional-order differential processing method. This meticulous signal processing method not only greatly improves the system's adaptability and response speed to changes in external working conditions, but also significantly enhances the overall industrial control accuracy, enabling the spraying robot to maintain efficient and stable operating performance in a changing 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 achieving more refined industrial control. This series of innovative measures work together to mark a major breakthrough in the control technology of hydraulically driven shotcrete robots, opening up a new path for improving the efficiency and quality of engineering operations.

[0054] (2) The present invention deeply integrates a variety of advanced signal processing technologies and intelligent control algorithms to construct a highly adaptive, self-learning and self-optimizing hydraulically driven shotcrete manipulator control system. At the signal processing level, the system adopts adaptive threshold noise reduction technology, combined with wavelet packet transform and sliding window normalization processing, which can dynamically identify and effectively suppress environmental noise and non-stationary interference, significantly improving the signal-to-noise ratio and feature extraction accuracy of the signal; at the same time, the cross-correlation analysis module monitors the coupling relationship between low-frequency and high-frequency signals in real time, promptly identifies abnormal coupling and implements frequency domain selective attenuation, thereby ensuring the physical consistency of signal processing and system stability. In terms of intelligent control, the system introduces a three-level cascade neural network architecture, combined with a dynamic weight adjustment mechanism and an incremental learning algorithm, which can adaptively optimize control parameters according to historical working conditions and real-time load status, realize closed-loop optimization from experience learning to online adjustment, and 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 effects, while the latter ensures the safety and feasibility of output instructions through a multi-level verification strategy, effectively preventing misoperation and system out of control. At the module maintenance level, the system has automatic calibration, real-time fault self-diagnosis, and data memory update functions. It can not only regularly calibrate the frequency response characteristics and phase consistency of the processing channel, but also quickly locate the source of the fault and enable redundant or estimation algorithms when an anomaly occurs, to ensure the reliability and stability of the system's long-term continuous operation. In summary, these integrated, intelligent, and adaptive technical means and mechanisms work together to enable the present invention to maintain an efficient, stable, and safe operating state when facing complex and changing operating environments, significantly improving the performance and intelligence level of the spraying robot in actual engineering applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention will be further described below with reference to the accompanying drawings.

[0056] Figure 1 This is a flow chart of an adaptive control method for a hydraulically driven shotcrete manipulator according to the present invention;

[0057] Figure 2 The present invention is a flowchart of an adaptive control system for a hydraulically driven shotcrete manipulator. DETAILED DESCRIPTION

[0058] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0059] See also Figure 1 As shown, the present invention is an adaptive control method for a hydraulically driven shotcrete manipulator, comprising the following steps:

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

[0061] S2: Perform frequency domain decomposition on the collected signal to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock;

[0062] S3: Performing dynamic weight adjustment processing on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes;

[0063] S4: performing phase compensation processing with adjustable differential order on the high-frequency characteristic signal to generate a second control characteristic value for suppressing pressure oscillation;

[0064] S5: Inputting the first control characteristic value and the second control characteristic value into the drive control model of the shotcrete manipulator to generate an adjustment instruction for the hydraulic actuator.

[0065] In S1, the pressure fluctuation signal, flow change signal and robot arm displacement signal of the hydraulic system are collected in real time, including:

[0066] The signal acquisition module, serving as the input to the system, is primarily responsible for acquiring key parameters during the operation of the hydraulic system in real time. This module contains three independent sensing units: a pressure sensor, a flow sensor, and a displacement sensor. Each unit is electrically isolated to prevent signal crosstalk. The pressure sensor uses a diffused silicon pressure sensor, installed at the inlet and outlet of the hydraulic cylinder. Its measurement range is 0-40 MPa, and its sampling frequency is set to 1 kHz, enabling it to accurately capture transient characteristics in pressure fluctuation signals. The sensor output signal is preprocessed with a second-order Butterworth low-pass filter, with a cutoff frequency set to 500 Hz, to eliminate high-frequency noise interference.

[0067] The flow sensing unit, a turbine flowmeter installed in the main oil line, achieves a measurement accuracy of ±0.5% FS. To accommodate the rapid flow rate fluctuations during shotcrete operations, a dual-sampling rate mechanism is used for flow signal acquisition: a 100Hz base sampling rate is used under steady-state conditions, automatically switching to a 1kHz high-speed sampling mode when the flow rate change exceeds a set threshold. The pulse signal output by the flowmeter is shaped by a Schmitt trigger and fed into a counter for processing, ultimately converting it into a standard analog signal.

[0068] The displacement sensing unit uses a magnetostrictive linear displacement sensor to measure the real-time position of each joint of the robotic arm. The sensor, installed inside the piston rod of the hydraulic cylinder, achieves a resolution of 0.01mm and an absolute measurement error of no more than ±0.1mm. To eliminate the effects of mechanical vibration on the measurement results, the displacement signal undergoes a moving average filter. The window length is adaptively adjusted based on the robotic arm's movement speed: a 20-point window is used to ensure measurement accuracy at low speeds, while a 5-point window is reduced at high speeds to minimize phase lag.

[0069] The output signals of the three sensing units are transmitted to the main controller via the CAN bus, with a fixed transmission cycle of 1ms. To improve system reliability, signal transmission adopts a dual-redundancy design: at each sampling moment, both the current value and the previous cycle value are transmitted simultaneously. When data anomalies are detected, historical data is automatically used. The main controller's receiving end has a data verification mechanism to eliminate and interpolate outliers that fall outside the acceptable range.

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

[0071] To adapt to varying operating conditions, the signal acquisition module supports online parameter configuration. Operators can adjust settings such as sampling frequency and filtering parameters for each channel through the human-machine interface. Modified parameters take effect immediately and are stored in non-volatile memory. The most recently valid configuration is automatically loaded upon system power-up, ensuring parameter setting continuity.

[0072] The signal acquisition module features self-diagnostic functionality, regularly checking the operating status of each sensor. If a sensor failure is detected, it automatically switches to a backup sensor or activates an estimation algorithm to maintain system operation, while displaying an alarm on the user interface. The diagnostics cover multiple indicators, including sensor supply voltage, signal amplitude range, and rate of change rationality, to ensure the reliability of the collected data.

[0073] In S2, the collected signal is subjected to frequency domain decomposition processing to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock, specifically including:

[0074] The frequency domain decomposition module is the core processing unit of this invention, primarily responsible for decomposing the acquired composite signal into low-frequency and high-frequency signatures with clear physical meaning. This module utilizes a hybrid signal processing architecture, combining traditional filtering techniques with modern time-frequency analysis methods, to adapt to the complex and changing working environment of the shotcrete manipulator. The module input receives the raw data stream from the signal acquisition module, first performing a signal quality check and marking and processing any missing or abnormal data segments.

[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 50-150Hz, with the default setting of 100Hz. The parameters of the filter are configured digitally, and a zero-phase filtering algorithm is used to avoid introducing time delay. The second-stage comb filter bank consists of eight parallel bandpass filters, with the center frequencies set at key frequency points from 1kHz to 5kHz. The bandwidth of each filter is set to 50Hz, which can effectively extract the characteristic harmonic components of the pressure shock. The output of the filter bank passes through a gain compensation circuit to ensure the amplitude consistency of the signals in each frequency band.

[0076] Sliding window normalization is implemented for low-frequency characteristic signals. This processing unit utilizes a circular buffer structure, with the window length automatically adjusted based on the robot's real-time motion speed: a long window of 500ms is used for speeds below 0.1m / s, a medium window of 200ms is used for speeds between 0.1-0.5m / s, and a short window of 50ms is used for speeds above 0.5m / s. Normalization involves two steps: DC component removal and amplitude normalization. The processed signal maintains its original phase information, while the amplitude is uniformly adjusted to the standard operating range.

[0077] The time-domain envelope detector reconstructs the amplitude of the high-frequency characteristic signal. This circuit utilizes precision full-wave rectification combined with a second-order low-pass filter, with a cutoff frequency set at 200Hz, to fully preserve the envelope characteristics of the impulse waveform. To eliminate residual carrier waves, an adaptive notch filter is incorporated into the circuit's post-processing stage, whose center frequency automatically tracks the dominant frequency component of the input signal. The reconstructed envelope signal passes through a sample-and-hold circuit, holding the impulse peak for an appropriate amount of time to facilitate subsequent processing modules in capturing transient characteristics.

[0078] The wavelet packet transform unit operates as an auxiliary processing channel. Using the DB4 wavelet basis function, this unit decomposes the input signal into 16 layers, generating detail coefficients for 16 frequency bands. Low-frequency reconstruction uses coefficients from layers 3 to 5, synthesizing a valid 0-100 Hz signal through an inverse wavelet transform. High-frequency reconstruction uses coefficients from layers 10 to 12, corresponding to a frequency range of 1 kHz to 4 kHz. Wavelet processing utilizes overlapping segmentation, with each segment consisting of 1024 points, and 256 points of overlap between segments to ensure time-domain continuity.

[0079] Adaptive threshold noise reduction is implemented on high-frequency signals reconstructed using wavelet transforms. This algorithm estimates the noise level in real time and dynamically adjusts the threshold curve based on the signal's power spectrum. Hard thresholding is used under steady-state conditions to preserve significant impulse components; under transient conditions, soft thresholding is used to smooth out small fluctuations. A 100ms threshold update cycle ensures timely tracking of changing operating conditions. The denoised signal undergoes phase correction to compensate for the group delay introduced by the wavelet transform.

[0080] The cross-correlation analysis module monitors the coupling relationship between low-frequency and high-frequency signals. This module calculates the normalized cross-correlation coefficient between the two signals using a 1-second time window and a 100-ms sliding step. When the correlation coefficient exceeds a threshold of 0.7, abnormal coupling is detected and a compensation mechanism is activated. 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 non-volatile memory, forming a continuously optimized knowledge base.

[0081] The signal output interface packages and transmits the processed low-frequency and high-frequency characteristic signals. Each data packet contains a time series of 100 sampling points, along with a timestamp and signal quality indicator. The transmission protocol supports error retransmission to ensure data integrity. The output interface also provides real-time monitoring signals for external devices to verify processing results.

[0082] The module features a built-in self-calibration function that regularly performs the following calibration process: first, injecting a standard test signal verifies the frequency response characteristics of each processing channel; then performing a white noise test to evaluate the system's noise suppression capabilities; and finally, performing a step response test to verify transient handling performance. Calibration results are automatically reported, and abnormalities trigger an alarm signal. The calibration cycle is configurable and defaults to every 24 hours.

[0083] In S3, dynamic weight adjustment processing is performed on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes, specifically including:

[0084] The dynamic weight adjustment module is one of the core processing units of the present invention, which is specifically responsible for intelligent weight allocation and feature fusion of low-frequency characteristic signals. The module adopts a three-level cascaded neural network architecture, and each level of the network has a specific feature extraction function. The first-level network is configured as a dual-input single-output structure, which specifically processes the pressure-flow coupling relationship of the hydraulic system. The network contains 32 processing nodes and uses a Sigmoid activation function, which can accurately capture the nonlinear dynamic characteristics between pressure and flow. The preprocessing unit at the network input end will standardize the original signal to eliminate the impact of dimensional differences.

[0085] The second-level network focuses on extracting features from the robot's motion trajectory. This network utilizes a time-series processing structure with an input window length of 50 sampling points, enabling it to identify dynamic features such as the robot's velocity and acceleration. The network incorporates long- and short-term memory units to memorize typical motion patterns. During feature extraction, the network automatically ignores minor fluctuations caused by measurement noise and focuses on identifying the true motion intent. The network output is connected to a feature selector to retain only the most representative motion features.

[0086] The third-level network serves as the feature fusion center, intelligently combining features extracted by the first two levels. This network adopts a fully connected structure and consists of three processing layers, with 64, 32, and 16 nodes in each layer, respectively. Dropout technology is used during network training to prevent overfitting, with a retention probability set to 0.7. The fused features are normalized and converted into a load status assessment value, a dimensionless parameter ranging from 0 to 1, with larger values ​​indicating heavier loads.

[0087] The weight coefficient calculation unit dynamically adjusts the importance weights of each characteristic channel based on the load state assessment value. This unit has a built-in nonlinear mapping function and uses a lookup table to achieve fast calculations. The weight calculation of the pressure characteristic channel uses an increasing curve, with a gradual increase in the light load range and a rapid increase in the heavy load range. The weight calculation of the displacement characteristic channel uses a decreasing curve, maintaining a high weight at low speeds and gradually decreasing at high speeds. The weight adjustment process introduces an inertia factor to prevent control oscillation caused by sudden changes in weights.

[0088] The core of the control variable generation module is the load-control variable conversion table. This table utilizes a three-dimensional structure, indexed by load state assessment, motion velocity, and pressure change rate. The table data is maintained through a combination of offline training and online learning. Initial values ​​are derived from historically optimal parameters, and continuous optimization is performed during runtime based on actual control performance. Table queries utilize a trilinear interpolation algorithm to ensure continuous and smooth output. At the end of each operating cycle, the module automatically records the optimal control parameters for incremental table updates.

[0089] The load signature memory library utilizes a circular buffer structure with a storage capacity of 1000 operating cycles. Each storage unit contains a complete feature vector, weight combination, and control effect evaluation. Data is automatically compressed during storage, preserving key features while saving storage space. The memory library maintenance process regularly organizes data, deleting excessively duplicated records and retaining the most representative operating patterns.

[0090] The pattern matching algorithm utilizes improved dynamic time warping technology, effectively processing work segments of varying lengths. The matching process considers the temporal morphology and statistical properties of feature vectors to calculate a comprehensive similarity score. The algorithm sets a matching quality threshold. When the optimal match score falls below the threshold, the algorithm automatically switches to a default weight combination to ensure system reliability. Matching results are accompanied by a confidence indicator for reference by subsequent processing modules.

[0091] The incremental learning mechanism uses a small-batch update strategy, with each adjustment not exceeding 10% of the original value. The learning process is subject to dual constraints: rapid adaptation to changing operating conditions while maintaining system stability. The direction of weight fine-tuning is determined by the control performance trends over the last 10 cycles, using a sliding average method to eliminate random interference. The learning algorithm incorporates a built-in forgetting factor to automatically reduce the influence of outdated data, ensuring that the system consistently tracks the current optimal operating point.

[0092] The first control eigenvalue generated by the module's output interface consists of two components: a static compensation component derived from the baseline value in the conversion table, and a dynamic adjustment component reflecting real-time learning results. The mixing ratio of the two components is automatically adjusted based on operating stability, prioritizing the static component during steady-state conditions and the dynamic component during transient conditions. The output data also includes a quality flag indicating the confidence level of the current control parameters.

[0093] In S4, a phase compensation process 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 in this invention, specifically responsible for processing high-frequency characteristic signals and generating control eigenvalues ​​to suppress pressure oscillations. This module utilizes innovative differential order adjustable technology, intelligently adjusting its processing strategy based on impact intensity. The module input receives the high-frequency characteristic signals from the frequency domain decomposition module and first performs a signal integrity check to ensure data validity. The input buffer utilizes a double-buffer design to ensure uninterrupted data flow processing.

[0095] The impact strength grading unit is a core component of this module. It monitors the amplitude characteristics of the input signal in real time and divides it into three operating ranges: the low amplitude range corresponds to 0-20% of the range, the medium amplitude range corresponds to 20-60% of the range, and the high amplitude range corresponds to 60-100% of the range. Each range is configured with an independent differential order parameter: the low range uses a shallow order of 0.3-0.5, the medium range uses a medium order of 0.5-0.7, and the high range uses a deep order of 0.7-0.9. A 5% hysteresis band is set when switching between ranges to prevent boundary oscillations. The grading results are updated in real time via the status register and are available for subsequent processing units.

[0096] The differential processor implements adjustable-order differential operations. It uses digital filtering to approximate fractional-order differentials, with its core algorithm based on a weighted moving average. During processing, the differential order is dynamically adjusted based on the impact intensity classification results, ensuring a smooth transition and avoiding abrupt changes. The processor also includes a built-in anti-aliasing filter whose cutoff frequency automatically matches the current differential order to ensure signal quality. The output of the differential operation undergoes amplitude normalization to eliminate gain fluctuations caused by order changes.

[0097] The delay compensation unit is responsible for correcting phase deviation in the signal processing chain. This unit utilizes an all-pass filter structure, and the phase compensation amount is programmable. Compensation parameters are obtained through offline calibration and stored in a calibration table. During real-time processing, the corresponding compensation parameters are automatically found based on the current differential order. After compensation, the signal phase error is controlled within ±5 degrees, meeting the control system's accuracy requirements. The unit also features a built-in self-test function for regular verification of phase characteristics.

[0098] The signal coupler intelligently fuses the processed differential signal with the original high-frequency features. The fusion algorithm utilizes conditional weighted multiplication, prioritizing trend information when the signal amplitude is small and enhancing fluctuations when the amplitude is large. The coupling coefficient dynamically adjusts based on the impact intensity to ensure the physical meaning of the output eigenvalues ​​is clear. The coupling process preserves the polarity characteristics of the original signal to avoid information distortion. A limiter is configured at the output to constrain the eigenvalues ​​within a safe range.

[0099] The pressure shock signature database serves as the foundation for this module's intelligent learning. The database utilizes a hierarchical storage structure, with the upper layer storing typical waveform features and the lower layer recording the corresponding optimal processing parameters. During data acquisition, 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 is designed to hold 1,000 records and employs a least recently used strategy for space management.

[0100] The real-time pattern recognition engine utilizes an improved dynamic time warping algorithm. It performs pattern matching every 50ms, calculating the similarity between the current signal and a database template. This matching process considers multiple metrics, including waveform morphology, spectral distribution, and statistical characteristics. To enhance real-time performance, the engine employs a hierarchical search strategy, performing a coarse screening followed by a finer matching process. Recognition results are accompanied by a confidence score, and only matches with a score above a threshold are considered.

[0101] A closed-loop verification mechanism forms the module's quality assurance system. This mechanism continuously monitors the effectiveness of suppressing pressure oscillations, using evaluation metrics such as peak decay rate, settling time, and overshoot. The verification cycle is 100ms, using sliding window statistics. If the performance falls short of expectations, a parameter reconfiguration process is triggered. This reconfiguration process uses a gradient heuristic to gradually adjust the differential order until satisfactory results are achieved. The results of each reconfiguration are fed back into the feature database to enable continuous optimization.

[0102] The module's output interface generates a second control characteristic value, a signed number that reflects both the direction of pressure change and the intensity of oscillation. The output data is accompanied by a quality flag, indicating the confidence level of the current processing. The interface protocol supports real-time monitoring, allowing external devices to read intermediate processing results. The module's operating status is intuitively displayed via indicator lights, including normal operation, learning update, and abnormality alarm.

[0103] In S5, the first control characteristic value and the second control characteristic value are input into the drive control model of the shotcrete manipulator to generate an adjustment instruction for the hydraulic actuator, specifically including:

[0104] The drive control module, the final output link of the present invention, is responsible for converting processed characteristic signals into executable hydraulic control instructions. This module utilizes a dual-channel hybrid control architecture, with the primary channel handling steady-state load compensation requirements and the secondary channel addressing dynamic pressure oscillations. During module initialization, the module first loads the default parameter configuration, including key parameters such as the base gain coefficient and clipping threshold. These parameters are stored in non-volatile memory to ensure they are not lost during power outages.

[0105] The dual-channel feature fusion module uses a weighted summation algorithm to achieve signal synthesis. The main channel is equipped with a fixed-gain amplifier, and the gain coefficient is set to the optimal value in the range of 0.7-0.9 to ensure the dominance of the basic control quantity. The auxiliary channel is equipped with an automatic gain adjuster, whose gain coefficient is dynamically calculated based on the real-time pressure fluctuation rate. The greater the fluctuation rate, the higher the gain, and the maximum does not exceed 0.5. The two signals are low-pass filtered before synthesis. The cutoff frequency of the main channel is set to 10Hz and that of the auxiliary channel is set to 50Hz, which ensures dynamic response while avoiding high-frequency interference. The synthesis operation uses a saturated adder to prevent output overflow.

[0106] The command smoothing processor utilizes a digital ramp-limiting algorithm. It calculates the gradient of control command changes in real time and automatically inserts transition command points when the instantaneous rate of change exceeds a set threshold. The threshold parameters are set based on the dynamic characteristics of the actuator and, for a typical hydraulic servo valve, are limited to a full-scale change rate of 5-10 times per second. The processor also includes a built-in prediction function, which uses command trends from the previous three cycles to predict the next demand and perform smoothing in advance. In exceptional cases, when urgent control needs are detected, the constraints can be temporarily relaxed to prioritize system safety.

[0107] The actuator characteristic compensation unit consists of two main functional blocks: a deadband compensator and a coupling demodulator. The deadband compensator uses a table lookup to predict the actual valve spool opening based on the current command value and historical movement direction. The compensation amount automatically adjusts over time to accommodate changes in valve component wear. The coupling demodulator monitors the system's pressure-flow relationship and injects a decoupling compensation signal when a strong coupling condition is identified. Compensation parameters are regularly updated using an online identification algorithm to ensure adaptability to characteristic variations under varying oil temperatures and operating pressures.

[0108] The control feature-execution effect mapping database is stored in a time-series data structure. Each record contains three pieces of information: the control feature vector, the output instruction, and the actual execution effect, with a time resolution of 10ms. The database incorporates an automatic organization mechanism to regularly merge similar records and delete redundant data. The query interface supports multi-dimensional indexing, enabling rapid retrieval of historically optimal control strategies. The database is designed to store the last 100 hours of operating data, using a first-in, first-out management strategy.

[0109] A feedforward-feedback composite control architecture enables precise regulation. The feedforward controller directly analyzes the current eigenvector and generates a reference command through fuzzy inference, keeping the response time within 5ms. The feedback regulator uses an incremental PID algorithm to fine-tune the control variable based on the execution deviation, with a regulation cycle of 20ms. The outputs of the two branches are weighted and combined in a synthesizer. The weighting coefficients are dynamically adjusted based on the operating conditions, emphasizing feedforward in steady-state conditions and feedback in transient conditions. Anti-saturation protection is provided at the synthesizer output to prevent integral saturation.

[0110] The instruction safety verification process implements a three-level protection. The first level performs a value range check to ensure that the instruction is within the hardware's permitted range; the second level verifies the rationality of the change trend to prevent abnormal jumps; and the third level evaluates the execution prediction effect to avoid dangerous operations. If any level of check fails, the instruction reconstruction process is immediately initiated. The reconstruction algorithm retains the control intent of the original instruction and generates a safe and feasible replacement instruction through methods such as proportional scaling or piecewise approximation. The reconstruction process is recorded in real time for fault analysis.

[0111] The module's output interface utilizes industry-standard protocols and supports both 4-20mA analog output and CAN bus digital output. The output driver features overcurrent protection to ensure actuator safety. The interface circuit utilizes optoelectronic isolation to enhance interference immunity. After each output cycle, the module automatically generates an operation log, recording key parameters and abnormal events. This log data is accessible through the debug interface.

[0112] Module maintenance features include automatic calibration and fault self-diagnosis. Automatic calibration is performed regularly, injecting test signals to verify the performance of each processing step. The fault diagnosis system monitors the module's operating status in real time, checking for signal continuity, processing timeliness, and logical rationality. Diagnostic results are categorized as caution, warning, and critical, each with corresponding action strategies. Detailed logs are maintained for all maintenance operations, enabling post-mortem analysis.

[0113] See also Figure 2 As shown, an adaptive control system for a hydraulically driven shotcrete manipulator comprises:

[0114] A signal acquisition module, which is used to collect pressure fluctuation signals, flow change signals and mechanical arm displacement signals of the hydraulic system in real time;

[0115] A frequency domain decomposition module is used to perform frequency domain decomposition processing on the collected signal to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock;

[0116] a dynamic weight adjustment module, configured to perform dynamic weight adjustment processing on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes;

[0117] a phase compensation processing module, the phase compensation processing module being used 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;

[0118] A drive control module inputs the first control characteristic value and the second control characteristic value into a drive control model of the shotcrete manipulator to generate an adjustment instruction for the hydraulic actuator.

[0119] Working Principle: This invention achieves precise control under complex working conditions through multimodal signal fusion and intelligent control technology. The method involves: first, high-precision sensors are used to collect the hydraulic system's pressure, flow, and manipulator displacement signals in real time. A frequency-domain decomposition technique combining multi-stage filtering and wavelet packet transform is used to separate the original signals into low-frequency characteristic signals reflecting load changes and high-frequency characteristic signals representing pressure shocks. A three-level neural network feature extraction architecture is constructed for the low-frequency signals. Load compensation control variables are generated through dynamic weight adjustment, where the weight coefficients are nonlinearly mapped according to the load state assessment value and optimized online by combining historical operating condition pattern matching. High-frequency signals are processed with adjustable differential order based on shock intensity grading. Oscillation suppression control variables are generated through phase compensation and signal coupling, with the differential order dynamically adjusted based on real-time pattern recognition results. Finally, a dual-channel feature fusion strategy is used to intelligently integrate the two control variables. After command smoothing and actuator characteristic compensation, hydraulic adjustment commands are output. The system innovatively incorporates a control feature-execution effect mapping database and a feedforward-feedback composite control architecture to continuously optimize control performance through closed-loop verification and parameter self-learning. This method effectively solves the technical problem of balancing load adaptability and oscillation suppression in traditional control, and significantly improves the stability and control accuracy of shotcrete operations.

[0120] The above is a detailed description of an embodiment of the present invention. However, the content described is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. An adaptive control method for a hydraulically driven shotcrete manipulator, characterized in that: The following steps are involved: S1: Real-time acquisition of pressure fluctuation signals, flow change signals and robotic arm displacement signals of the hydraulic system; S2: Perform frequency domain decomposition on the collected signal to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock; S3: Performing dynamic weight adjustment processing on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes; S4: performing phase compensation processing with adjustable differential order on the high-frequency characteristic signal to generate a second control characteristic value for suppressing pressure oscillation; S5: Inputting the first control characteristic value and the second control characteristic value into the drive control model of the shotcrete manipulator to generate an adjustment instruction for the hydraulic actuator.

2. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The frequency domain decomposition process includes the following specific steps: A signal separation device based on a multi-stage filter was constructed. The first stage used a Butterworth filter with an adjustable cutoff frequency to extract the 0-100 Hz frequency band signal as the low-frequency characteristic signal. The second stage used a comb filter bank to extract the harmonic components of a specific frequency greater than 1 kHz as the high-frequency characteristic signal. The low-frequency characteristic signal is subjected to sliding window normalization processing, the window length is dynamically adjusted according to the movement speed of the robot arm, and the normalized signal is sent to the subsequent processing module; Amplitude reconstruction based on time domain envelope detection is performed on the high-frequency characteristic signal to retain the time-varying characteristics of the impulse waveform and eliminate the interference of the high-frequency carrier component.

3. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The frequency domain decomposition process further includes the following optimization steps: The original signal is decomposed into 16 layers using wavelet packet transform, and the detail coefficients of the 3rd to 5th layers are selected to reconstruct the low-frequency feature signal, and the detail coefficients of the 10th to 12th layers are selected to reconstruct the high-frequency feature signal; Adaptive threshold noise reduction is performed on the reconstructed high-frequency characteristic signal, and the threshold value is dynamically adjusted according to the signal-to-noise ratio under the current working conditions; A cross-correlation function between the low-frequency characteristic signal and the high-frequency characteristic signal is established. When the correlation coefficient exceeds the set threshold, the signal compensation mechanism is activated to correct the decomposition result.

4. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The dynamic weight adjustment process includes the following specific steps: A feature extraction network with three hidden layers was constructed. The first-level network extracted the pressure-flow coupling characteristics of the hydraulic system, the second-level network extracted the robot arm motion trajectory characteristics, and the third-level network integrated the first two levels of features to generate the load state evaluation value. According to the load state evaluation value, a weight coefficient of each characteristic channel is calculated by a nonlinear mapping function, wherein the weight of the pressure characteristic channel increases as the load increases, and the weight of the displacement characteristic channel decreases as the movement speed increases; The weighted multi-channel characteristics are input to the control quantity generation module, and the first control characteristic value is output through a preset load-control quantity conversion relationship table. The conversion relationship table is periodically updated according to historical operating condition data.

5. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The dynamic weight adjustment process further includes the following optimization steps: Establish a load feature memory library based on a sliding time window to store the characteristic signal change trajectory of the latest N sampling periods; The pattern matching algorithm is used to compare the current characteristic signal with the typical working condition patterns in the memory library, and the weight combination corresponding to the historical pattern with the highest matching degree is selected as the initial value; The initial weights are fine-tuned online through an incremental learning method, with the adjustment amplitude being proportional to the degree to which the current operating condition deviates from the typical mode. The first control characteristic value finally generated includes a static compensation component and a dynamic adjustment component.

6. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The differential order adjustable phase compensation process includes the following specific steps: A differential order adjustment mechanism based on shock intensity grading is established, dividing the amplitude of the high-frequency characteristic signal into three intensity ranges, corresponding to different differential order configurations: shallow order differentials are used in the low amplitude range to maintain system stability, medium order differentials are used in the medium amplitude range to achieve fast response, and deep order differentials are used in the high amplitude range for strong suppression. Perform phase calibration on the differential signal after order adjustment, and eliminate the phase lag introduced by the signal processing link through the delay compensation unit; The phase-calibrated differential signal is multiplied and coupled with the original high-frequency characteristic signal to generate a second control characteristic value with a pressure oscillation suppression effect. The characteristic value contains both pressure change trend information and instantaneous fluctuation intensity information.

7. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The differential order adjustable phase compensation process further includes the following optimization steps: Construct a pressure shock feature database to record the waveform patterns of high-frequency characteristic signals under different working conditions and their corresponding optimal differential order parameters; Use real-time pattern recognition technology to match the current high-frequency characteristic signal with the typical waveform in the characteristic database, and dynamically select the differential order combination based on the matching results; The generated second control eigenvalue is evaluated through a closed-loop verification mechanism. When the pressure oscillation suppression effect does not meet expectations, the differential order reconfiguration process is automatically triggered until a stable pressure suppression effect is achieved.

8. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The drive control model processing includes the following specific steps: Construct a dual-channel feature fusion module, set the main action channel of the first control eigenvalue and the auxiliary correction channel of the second control eigenvalue, 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 limiting on the fused control command, limiting the command change rate to not exceed the dynamic response capability of the hydraulic actuator; The generated adjustment instruction is corrected nonlinearly by the actuator characteristic compensation unit to compensate for the dead zone characteristics of the hydraulic valve and the coupling effect of flow and pressure, and the final executable mechanism drive instruction is output.

9. The adaptive control method for a hydraulically driven shotcrete manipulator according to claim 1, characterized in that: The drive control model processing also includes the following optimization steps: Establish a mapping database between control characteristics and execution effects, and store the corresponding relationship between historical control characteristic values ​​and actual execution effects; Adopting a feedforward-feedback composite control architecture, the feedforward branch directly generates a benchmark instruction based on the current control characteristic value, and the feedback branch makes fine adjustments based on the execution effect deviation; An instruction safety verification link is set up. When the generated adjustment instruction exceeds the working range of the actuator, the instruction reconstruction mechanism is activated to automatically adjust the feature fusion ratio while maintaining the control effect.

10. An adaptive control system for a hydraulically driven shotcrete manipulator, characterized in that: An adaptive control method for a hydraulically driven shotcrete manipulator according to any one of claims 1 to 9, comprising: A signal acquisition module, which is used to collect pressure fluctuation signals, flow change signals and mechanical arm displacement signals of the hydraulic system in real time; A frequency domain decomposition module is used to perform frequency domain decomposition processing on the collected signal to separate the low-frequency characteristic signal reflecting the load change and the high-frequency characteristic signal reflecting the pressure shock; a dynamic weight adjustment module, configured to perform dynamic weight adjustment processing on the low-frequency characteristic signal to generate a first control characteristic value for compensating for load changes; a phase compensation processing module, the phase compensation processing module being used 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 manipulator to generate an adjustment instruction for the hydraulic actuator.

Images