Cantilever beam forming machine multi-source data fusion control and intelligent monitoring method based on digital twinning

Abstract

The application relates to the technical field of intelligent control of bridge construction, and discloses a cantilever beam builder multi-source data fusion control and intelligent monitoring method based on digital twinning, which comprises the following steps: collecting the displacement, pressure and construction vibration intensity of each oil cylinder in real time; adopting attenuated memory improved extended Kalman filtering with unknown input decoupling estimation to jointly estimate the oil cylinder displacement, speed, pressure and unknown disturbance force, and outputting a multi-dimensional fusion feature vector containing fusion state estimation value, unknown disturbance force estimation value, health index and displacement estimation standard deviation; adopting an improved isolated forest based on a segmented cumulative probability density function to perform abnormal scoring and early warning; performing multi-cylinder load balancing adjustment and PID synchronous control based on the unknown disturbance force estimation value; and mapping the estimation result, abnormal score and early warning level to a digital twinning platform in real time, driving a three-dimensional model to perform state synchronization and visualized monitoring. Therefore, the multi-oil-cylinder collaborative control precision and abnormal detection capability of the cantilever beam builder are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for bridge construction, specifically to a method for multi-source data fusion control and intelligent monitoring of cantilever beam-building machines based on digital twins. Background Technology

[0002] Cantilever beam erecting machines are used in the construction of large bridges such as high-speed railways and urban rail transit systems to erect precast box girders. Their subsystems, including walking, lifting, side formwork, bottom basket, and rear anchoring, operate in coordination, involving the real-time acquisition and fusion processing of data from multiple types of sensors, such as hydraulic, displacement, and attitude sensors. Taking the lifting system as an example, a typical configuration involves eight hydraulic cylinders (e.g., 120 mm stroke, 25 MPa rated pressure, 50 t capacity per cylinder) synchronously lifting the precast box girder. Controlling the synchronization error and pressure difference between the cylinders is a core technical challenge to ensure construction safety. Existing technologies have the following main problems: (I) Limitations of Standard Extended Kalman Filter (EKF) in Multi-Cylinder Synchronous Control: During the lifting and traveling process of a cantilever beam-building machine, factors such as beam eccentric loading and uneven foundation settlement can generate unknown disturbance forces on the hydraulic cylinders that are difficult to quantify in advance. Existing EKF methods generally assume that the system excitation is completely known, making it impossible to effectively estimate the aforementioned unknown disturbance forces, resulting in insufficient accuracy of multi-cylinder synchronous error control. In addition, hydraulic shock and construction vibration can cause sudden changes in sensor measurement noise within a short period of time. Standard EKF uses equal weighting for historical measurement statistics, and when contaminated historical measurements accumulate, they will continuously skew the current state estimation. At the same time, hydraulic cylinder sealing friction, mechanical clearance, and elastic deformation cause the system to exhibit nonlinear characteristics. The linearization approximation of standard EKF will introduce a non-negligible model mismatch error during large-stroke lifting. Since the degree of influence of eccentric loading varies among cylinders, the unknown disturbance forces are coupled between cylinders, further increasing the error of independent estimation for each cylinder.

[0003] (II) Limitations of the Standard Isolation Forest Algorithm in Multi-Cylinder Collaborative Monitoring: Under normal construction conditions, the multi-cylinder hydraulic cylinders of a beam-building machine form a typical multi-dense cluster data distribution—the pressure and displacement feature points of each cylinder are highly concentrated in the normal operating range, while progressive anomalies (such as progressive pressure drift caused by slow aging of seals, and cumulative travel deviation caused by local deformation of the track) often appear in the gaps between normal data clusters. The standard isolation forest algorithm generates cut values ​​randomly with a uniform distribution, which is insufficient for isolating such anomalies between dense clusters and is prone to missed detections. In addition, safety boundary anomalies such as cylinder pressure exceeding limits and displacement exceeding limits usually appear at the ends of the feature space. The inherent bias of the standard isolation forest algorithm, which is "high-resolution concentration towards the edge," actually weakens the detection sensitivity of safety boundary anomalies at the ends.

[0004] (III) Coordinated handling of unknown disturbance forces and multi-cylinder synchronization anomalies: In existing technologies, state estimation and anomaly detection are usually operated as independent modules. When a cylinder has a latent fault, the additional disturbance force it generates will affect the state estimation accuracy of adjacent cylinders through coupling transmission in the hydraulic system. Anomaly detection based solely on the original sensor signal cannot fully utilize the physical state information revealed by the estimation model, making it difficult to effectively separate early fault signals from normal construction fluctuations. Summary of the Invention

[0005] The purpose of this invention is to address the problems encountered by cantilever beam erecting machines during lifting, traveling, and attitude adjustment processes, such as asynchronous operation of multiple hydraulic cylinders, difficulty in quantifying off-center load disturbances, significant nonlinearity of the hydraulic system, strong vibration and noise interference from the construction environment, and difficulty in early identification of abnormal signs. This invention proposes a multi-source data fusion control and intelligent monitoring method for cantilever beam erecting machines based on digital twins. This method involves the fusion processing of multi-source heterogeneous sensor data during cantilever beam erecting machine operation, collaborative state estimation of multiple hydraulic cylinders, progressive anomaly detection, and digital twin collaborative monitoring. This invention aims to achieve joint estimation of cylinder displacement, pressure, and unknown disturbance forces by improving extended Kalman filtering, and to achieve hierarchical early warning of progressive anomalies and boundary over-limit anomalies by improving isolated forest filtering. The estimation results are then linked with the anomaly detection results for multi-cylinder load balancing, synchronous control, and digital twin visualization monitoring, thereby improving the state perception capability, anomaly identification capability, and control coordination of the cantilever beam erecting machine during construction.

[0006] In a first aspect, the present invention provides a multi-source data fusion control and intelligent monitoring method for cantilever beam-building machines based on digital twins, including: S1: Real-time acquisition of displacement measurement values, pressure measurement values, and construction vibration intensity parameters of each cylinder of the cantilever beam building machine; S2: For each hydraulic cylinder, using the displacement and pressure measurements as measurement inputs, a recursive filtering process including nonlinear state prediction and measurement updates is executed. The attenuation memory weights of the filter are dynamically adjusted using the construction vibration intensity parameters, and an improved extended Kalman filter with attenuation memory and unknown input decoupling estimation is used to perform joint state estimation of the cylinder's displacement, velocity, pressure, and unknown disturbance force. The output includes a multi-dimensional fusion feature vector containing the fused state estimate, the unknown disturbance force estimate, the health index calculated based on the covariance matrix, and the standard deviation of the displacement estimate. S3: Using the multidimensional fusion feature vector as input, an anomaly score is performed using an improved isolated forest based on a piecewise cumulative probability density function, and the anomaly score and its corresponding warning level are output. S4: Based on the estimated value of the unknown disturbance force, perform multi-cylinder load balancing adjustment and PID synchronous control to generate target control commands for each cylinder in order to correct the multi-cylinder synchronization error; S5: The fusion state estimate, unknown disturbance force estimate, displacement estimate standard deviation, health index, anomaly score and early warning level are mapped to the digital twin platform in real time through a standardized data protocol, driving the three-dimensional twin model to perform state synchronization and visual monitoring.

[0007] Furthermore, in step S2: the nonlinear state prediction specifically includes: using a nonlinear state transition function to predict the joint state vector of the hydraulic cylinder, the nonlinear state transition function being used to characterize the dynamic characteristics of the hydraulic cylinder, including nonlinear friction force. The attenuation memory improved extended Kalman filter with unknown input decoupling estimation specifically includes: After the nonlinear state prediction is completed at each time step, the estimated value of the unknown disturbance force is calculated in real time using the least squares method based on the measurement residuals, thereby achieving online decoupling between the unknown disturbance force and the system state. The calculation formula is as follows: ; in, The unknown perturbation force at time k+1 The estimated value, The unknown input influence matrix is ​​determined by the force geometry of the hydraulic cylinder. This is the observation vector containing displacement and pressure measurements at time k+1. For nonlinear observation functions, The state vector at time k+1 The prior estimate of .

[0008] Furthermore, in step S2, the attenuation memory weight of the filter dynamically adjusted using the construction vibration intensity parameter adopts an attenuation memory weight mechanism, specifically including: Based on the construction vibration intensity parameters at the current moment Vibration intensity compared to normal construction reference The ratio of the decay memory weight coefficient is dynamically calculated. and the decay memory weight coefficient In the recursive process of introducing the prediction error covariance matrix, the influence weight of historical measurement data on the current state estimation is reduced when vibration intensifies. The formula is as follows: ; in, To estimate the error covariance matrix a priori, Let Jacobian be the state transition matrix. For the posterior estimation of the error covariance matrix, Let be the process noise covariance matrix.

[0009] Furthermore, in step S2, the joint state estimation also employs Chebyshev polynomials to perform online modeling of the nonlinear frictional force of the hydraulic cylinder, specifically including: Displace the hydraulic cylinder Normalize to interval [ After [1,1], the normalized cylinder displacement is obtained. Then expressed as a weighted sum of Chebyshev polynomials: ; in, for Chebyshev polynomials The coefficients to be identified are... For the nonlinear frictional force of the hydraulic cylinder, Let be the order of the Chebyshev polynomial.

[0010] Furthermore, step S2 also includes a multi-cylinder coupled disturbance broadcast compensation mechanism, specifically including: When the The estimated value of the unknown disturbance force of the cylinder satisfies The broadcast is triggered at a certain time, in which, The preset threshold coefficient, This is the rated reference disturbance force; After the broadcast is triggered, the estimated value of the unknown disturbance force of the j-th cylinder is corrected to obtain the value of the j-th cylinder. The cylinder is Estimated value of unknown disturbance force at time , used for the Correction of cylinder state prediction for the next moment: ; in, The off-center load coupling coefficient was pre-calibrated through single-cylinder loading tests during the commissioning phase, and .

[0011] Furthermore, in step S3, the improved isolated forest cut value generation method based on the piecewise cumulative probability density function includes: When splitting at each node of the isolated tree, the probability of each interval being selected is determined along the selected feature axis according to the power function of the length of each data interval, so that the cut value falls preferentially in the sparse gap region of the data points; after determining the selected interval, the cut value is randomly generated in a uniform distribution within the selected interval.

[0012] Furthermore, in step S3, the improved isolated forest based on the piecewise cumulative probability density function also includes a U-shaped probability density compensation mechanism: During the generation of random numbers for node splitting, a U-shaped probability density correction is superimposed to increase the probability that the cut value falls in the end region of the feature space, thereby enhancing the detection sensitivity for safety boundary anomalies such as pressure exceeding limits or travel exceeding limits.

[0013] Furthermore, in step S3, the improved isolated forest based on the piecewise cumulative probability density function uses the bias-enhanced anomaly scoring function to calculate the final anomaly score. The deviation enhancement anomaly scoring function includes both path score and deviation score. The path score reflects the ease with which a sample is isolated, and the deviation score reflects the degree to which the sample feature value deviates from the normalized average of the splitting threshold of each node. The final anomaly score is the product of the path score and the deviation score.

[0014] Furthermore, the health index is defined as: the mean of the steady-state covariance matrix of the i-th cylinder under normal operating conditions during the commissioning period. traces and current Posterior covariance matrix at time 1 The ratio of traces: ; in, For the first The cylinder is Health index at any time Let be the trace of the matrix.

[0015] Furthermore, in step S4, the multi-cylinder load balancing adjustment specifically includes: calculate Arithmetic mean of the estimated unknown disturbance forces of all cylinders at any given time ; According to the arithmetic mean With the The cylinder is Estimated value of unknown disturbance force at time The deviation is calculated as follows: Cylinder pressure correction: ;in, For load balancing gain, The piston area; The pressure correction amount Superimposed on the system reference pressure command Above, generate the first The cylinder's ultimate target pressure ; The final target pressure shall not exceed the alarm limit specified in the equipment specification. If it exceeds the limit, the alarm will be truncated to the alarm limit and a level 2 warning will be triggered.

[0016] Secondly, embodiments of the present invention also provide an electronic device, the electronic device comprising: One or more processors; Storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the multi-source data fusion control and intelligent monitoring method for cantilever beam construction machines based on digital twins as described in any embodiment of the present invention.

[0017] Thirdly, embodiments of the present invention also provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the multi-source data fusion control and intelligent monitoring method for cantilever beam-building machines based on digital twins as described in any embodiment of the present invention.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention can jointly estimate the displacement, pressure and unknown disturbance force in the multi-cylinder system of the cantilever beam machine, so that the control system can not only obtain the apparent sensing data, but also obtain the state information related to the off-center load and coupled disturbance, providing a more stable feedback basis for subsequent control.

[0019] (2) By introducing a vibration intensity-based attenuation memory mechanism, this invention can appropriately reduce the impact of historical errors on the current estimate when construction vibration is enhanced or sensor measurements are contaminated, which helps to reduce the interference of short-term noise on state estimation.

[0020] (3) The present invention can use a multi-cylinder coupling broadcast compensation mechanism to coordinate the correction of the disturbance force information of a certain cylinder to other cylinders, which helps to improve the problem of off-center load diffusion in the multi-cylinder synchronization process and improve the consistency of multi-cylinder action.

[0021] (4) This invention improves the ability to identify progressive anomalies, boundary exceedance anomalies, and latent fault signs by changing the selection of isolated forest cutting intervals to a method more suitable for dense cluster distribution and introducing boundary enhancement and deviation enhancement scoring mechanisms, thereby reducing missed detections caused by relying solely on original sensor fluctuations. This anomaly detection method has been verified in typical faults such as track jamming, sensor drift, and lateral deviation exceedance in the traveling cylinder, with a significant improvement in detection rate.

[0022] (5) By setting multi-level early warning thresholds, the present invention can classify abnormal states into different levels such as monitoring, alarm and shutdown, so that the system can take different handling strategies according to the severity of the abnormality, thereby improving the level and timeliness of construction safety management.

[0023] (6) The present invention uses a digital twin platform to synchronously map the fusion status, anomaly score, early warning level and control results, which can realize three-dimensional visualization monitoring and historical traceability of the construction process, making it easy for operators to intuitively grasp the equipment operating status and anomaly evolution trend.

[0024] In summary, this invention integrates state estimation, anomaly detection, load balancing control, and digital twin monitoring into a unified closed loop, which can better adapt to the real-time control requirements of cantilever beam-making machines under complex working conditions and provide data support for subsequent maintenance decisions. Attached Figure Description

[0025] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 This is a flowchart illustrating a multi-source data fusion control and intelligent monitoring method for a cantilever beam-building machine based on digital twins, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the system architecture provided in an embodiment of the present invention; Figure 3 This is a flowchart of the AEKF-UI single cylinder state estimation provided in an embodiment of the present invention; Figure 4 This is a flowchart of the PCPD-iForest anomaly detection process provided in an embodiment of the present invention; Figure 5 This is the rear anchor lifting outrigger jacking control page (including 3D model schematic diagram) provided in the embodiment of the present invention. Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0026] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0027] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as being processed sequentially, many of these operations (or steps) may be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations may be rearranged. The process may be terminated when its operation is completed, but may also have additional steps not included in the figures. The process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0028] Example 1: like Figure 1 and Figure 2 As shown, a multi-source data fusion control and intelligent monitoring method for cantilever beam-building machines based on digital twins is proposed. The core innovations focus on two improved algorithms: an improved extended Kalman filter with decoupled estimation of unknown inputs using decay memory (AEKF-UI) and an improved isolated forest anomaly detection method based on piecewise cumulative probability density function (PCPD-iForest), and their collaborative integration in multi-cylinder synchronous control scenarios. Specifically, the method includes the following steps: S1: Real-time acquisition of displacement measurement values, pressure measurement values, and construction vibration intensity parameters of each cylinder of the cantilever beam building machine; S2: For each hydraulic cylinder, using the displacement and pressure measurements as measurement inputs, a recursive filtering process including nonlinear state prediction and measurement updates is executed. The attenuation memory weights of the filter are dynamically adjusted using the construction vibration intensity parameters, and an improved extended Kalman filter with attenuation memory and unknown input decoupling estimation is used to perform joint state estimation of the cylinder's displacement, velocity, pressure, and unknown disturbance force. The output includes a multi-dimensional fusion feature vector containing the fused state estimate, the unknown disturbance force estimate, the health index calculated based on the covariance matrix, and the standard deviation of the displacement estimate. like Figure 3 The diagram shown is a flowchart of the AEKF-UI single-cylinder state estimation provided in an embodiment of the present invention. The design of the attenuation memory improved extended Kalman filter (AEKF-UI) with unknown input decoupling estimation is as follows: S2.1 State-space description: The following state-space model is established for each hydraulic cylinder.

[0029] 1) State equations (nonlinear form): Nonlinear state prediction specifically includes: predicting the joint state vector of the hydraulic cylinder using a nonlinear state transition function. This nonlinear state transition function characterizes the dynamic characteristics of the hydraulic cylinder, including nonlinear frictional forces. The specific formula for nonlinear state prediction is as follows: (1) Among them, the joint state vector The parameters are, in order, cylinder displacement, velocity, hydraulic chamber pressure, and Chebyshev polynomial coefficients used to characterize nonlinear friction. Known control input (pump station flow command); The unknown disturbance force vector includes the additional hydraulic pressure generated by the beam eccentric loading and the multi-cylinder coupling force component. , These are process noise and measurement noise, initially set to zero-mean Gaussian white noise. It is a nonlinear state transition function that comprehensively describes the dynamics of the hydraulic cylinder (including nonlinear friction).

[0030] 2) Observation equation (nonlinear form): (2) Among them, observations Includes displacement sensor output and pressure sensor output. The function is a nonlinear observation function, considering the effect of changes in the hydraulic cylinder cavity volume on the pressure.

[0031] State transition Jacobian matrix and the observed Jacobian matrix They are respectively and The linearized approximation at the current operating point: (3) (4) in, and It is used only for covariance propagation and Kalman gain calculation in EKF and does not replace the nonlinear state transition and observation relationship in Equations (1) and (2). The state transition matrix is ​​in The Jacobian matrix at time step (linearized approximation) is given by the state function. For the state vector Taking the partial derivative yields the result; : Observation matrix in The Jacobian matrix at time (linearized approximation) is given by the observation function. For the state vector Taking the partial derivative yields the result; : Nonlinear state transition function, which comprehensively describes the dynamics of the hydraulic cylinder (including nonlinear friction force). : Nonlinear observation function, considering the effect of changes in the hydraulic cylinder cavity volume on pressure; : Joint state vector, including cylinder displacement, velocity, pressure and Chebyshev polynomial coefficients; : Time-state vector The posterior estimate; : Time-state vector The prior estimate (predicted value).

[0032] S2.2 Online decoupling estimation of unknown disturbance forces: At each time step, after the nonlinear state prediction of equation (1) is completed, the unknown perturbation force... The measurement residuals are calculated in real time using the least squares method to obtain... Unknown disturbance force at any time The estimated value This achieves online decoupling between unknown disturbance forces and system state. The calculation formula is as follows: (5) in, The unknown input influence matrix is ​​determined by the geometric relationship of the forces acting on the hydraulic cylinder. : Unknown disturbance force at any time The estimated value; : The unknown disturbance force vector at any given time includes the additional hydraulic pressure generated by the eccentric loading of the beam and the coupling force component of the multi-cylinder; : The observation vector at any given time includes the outputs of the displacement sensor and the pressure sensor.

[0033] This step is performed after the state prediction is completed and before the gain is updated, and extracts the off-center load moment and the additional force caused by foundation settlement from the measurement residual in real time.

[0034] Complete filter update steps: (6) (7) (8) in, : Kalman gain matrix at time step; : The prior estimation error covariance matrix (predicted value) at time 1; : Time observation matrix transpose; : Observe noise at all times The covariance matrix; Identity matrix, dimensions, and state vector same; : The posterior estimation error covariance matrix at time t (updated value).

[0035] S2.3, Decaying Memory Weight Mechanism and Adaptive Value Selection Rules: In step S2.3, the attenuation memory weight of the filter is dynamically adjusted using the construction vibration intensity parameter, employing an attenuation memory weight mechanism, specifically including: Based on the construction vibration intensity parameters at the current moment Vibration intensity compared to normal construction reference The ratio of the decay memory weight coefficient is dynamically calculated. and decay memory weight coefficient In the recursive process of introducing the prediction error covariance matrix, the influence weight of historical measurement data on the current state estimation is reduced when vibration intensifies. The formula is as follows: (9) in, : The posterior estimation error covariance matrix at time t; State transition matrix transpose; : Time-of-flight noise The covariance matrix; This is the prior estimate of the error covariance matrix. Decaying memory weighting coefficient, value range It is used to apply dynamic discounts to historical measurements in the recursion of the prediction error covariance matrix.

[0036] Real-time calculation based on the current construction vibration intensity: (10) in, The root mean square of the vibration acceleration at the current moment is calculated in real time by the vibration sensor; : Root mean square statistical value of vibration acceleration under normal construction reference conditions during the commissioning phase; Mapping sensitivity gain, used to adjust the response of vibration intensity changes to... The extent of the impact.

[0037] Mathematical verification: When When, inside the parentheses ,but (Historical measurements are nearly equally weighted, close to the standard EKF); when hour, ,but (Maximum discount for historical pollution measurement); When The upper time limit is 1.05, which is in line with the literature recommendations. scope.

[0038] The physical meaning and setting method: Characterizing the vibration intensity increment For mapping sensitivity, it is recommended to initialize with a value of 0. After the first construction phase was completed, a comparison was made. The standard deviation of the EKF measurement residuals can be set at different values ​​(the smaller the residual, the better the filtering effect), allowing for one-time tuning. This is particularly useful for construction environments with more severe vibrations (such as bridge construction in karst areas). The value can be appropriately increased to 1.5~2.0 to increase the discount for historical measurements; Data was obtained through vibration sensor statistics under normal construction benchmark conditions during the equipment commissioning phase.

[0039] S2.4 Modeling of Polynomial Nonlinear Hydraulic Systems: In step S2.4, the nonlinear frictional force of the hydraulic cylinder Nonparametric modeling using Chebyshev polynomials as basis functions: Displace the hydraulic cylinder Before substituting into the Chebyshev polynomial, interval normalization mapping is required. Specifically, the cylinder displacement is... Normalize to interval [ After [1,1], the normalized cylinder displacement is obtained. : (11) in, and The lower and upper limits of the cylinder stroke (e.g., lifting cylinder) Walking cylinder ); : Cylinder displacement; The normalized cylinder displacement is mapped to the interval [-1, 1].

[0040] After normalization, the nonlinear friction force can be expressed as a weighted sum of Chebyshev polynomials: (12) in, Nonlinear frictional force in hydraulic cylinders; : The Chebyshev polynomial is defined on the interval [-1, 1] and has both orthogonality and numerical stability guaranteed. The coefficients of the Chebyshev polynomial to be identified are jointly identified online recursively as part of the extended state vector, without the need for prior mechanical calibration. The order of the Chebyshev polynomial is usually taken as 3 to 5, and the initial estimation error tolerance is recommended not to exceed 30%.

[0041] S2.5 Definition of Hydraulic Cylinder Health Index: Specifically, the health index is defined as: the mean of the steady-state covariance matrix of the i-th cylinder under normal operating conditions during the commissioning period. traces and current Posterior covariance matrix at time 1 The ratio of traces: (13) in, : No. The cylinder is Health index at any time; The trace of a matrix (the sum of its diagonal elements); : No. The mean steady-state covariance matrix of the cylinder after EKF convergence under normal operating conditions during the commissioning period was obtained through statistical analysis of multiple complete construction cycles. : No. The cylinder is The posterior covariance matrix at time t.

[0042] Under normal operating conditions The health index approaches 1; when an abnormality occurs in the hydraulic cylinder, leading to an increase in the uncertainty of the state estimation, Increased size, health index drops below 1. Recommendation: The specific threshold for triggering a detailed inspection can be adjusted based on actual equipment measurement data.

[0043] S2.6 Multi-cylinder coupled disturbance broadcast compensation mechanism: The AEKF-UI of multiple hydraulic cylinders operates in parallel within the same control cycle. The specific multi-cylinder coupling disturbance broadcast compensation mechanism is as follows: (a) Triggering condition: When the first cylinder satisfies A broadcast is triggered at a certain time. : No. The cylinder is The estimated value of the unknown disturbance force at time; The initial value of the rated reference disturbance force is set at 5% of the rated bearing capacity of the cylinder (engineering basis: for the maximum eccentric load within the beam eccentric load limit, its additional force on a single cylinder is about 5% of the rated bearing capacity. This ratio is based on the engineering estimate of the beam eccentric load limit in the beam construction machine design specification). : Broadcast trigger threshold coefficient, the default value is 0.1. For light off-center load sensitive scenarios, a smaller value (such as 0.05) can be used, and for scenarios with greater noise interference, a larger value (such as 0.15) can be used. In actual use, it can be adjusted in the range of 0.05~0.20 according to the off-center load detection sensitivity requirements.

[0044] (b) Broadcast correction method: for the first cylinder The unknown perturbation force estimate at its current moment is superimposed from the first... Cylinder coupling correction amount , obtained the The cylinder is Estimated value of unknown disturbance force at time , used for the Correction of cylinder state prediction for the next moment: (14) in, : No. The cylinder is The estimated value of the unknown disturbance force at time; : No. The cylinder is The estimated value of the unknown disturbance force after broadcast correction at any given moment; : No. cylinder and the first Off-center coupling coefficient between cylinders.

[0045] Revised For the Correction of the cylinder's state prediction at the next moment The matrix itself remains unchanged (it is determined by fixed geometric relations).

[0046] Broadcast corrections only apply to the perturbation components of the state vector (state prediction corrections) and do not directly modify the covariance matrix. The covariance matrix is ​​still recursively derived according to equation (9), reflecting the uncertainty assessment of the cylinder state by AEKF-UI itself; the effect of the broadcast correction will be indirectly reflected in the covariance convergence process through the measurement updates at subsequent times (equations (7) and (8)).

[0047] (c) Calibration method: Coupling coefficient matrix The single-cylinder loading test during the commissioning phase determined that—for the first… The cylinder is subjected to a known additional load. (While keeping other cylinder control commands unchanged), record the first... Response change of cylinder pressure sensor (Converted to equivalent value), then: (15) in, During the single-cylinder loading test in the commissioning phase, the first... The known additional load applied by the cylinder; During the single-cylinder loading test in the commissioning phase, the first... The change in response of the cylinder pressure sensor (converted to equivalent force).

[0048] For a symmetrical 8-cylinder arrangement, typically only three typical positional relationships need to be defined: adjacent cylinders, spaced cylinders, and diagonal cylinders. The remaining cylinder pairs can be deduced from geometric symmetry. (The last sentence appears to be a separate, unrelated statement: "Symmetrically arranged adjacent cylinders...") The reference value range is 0.05~0.15, and the reference value range for asymmetrical or long-distance cylinder pairs is 0~0.05.

[0049] (d) Broadcast direction: Broadcast is multi-directionally triggered - any cylinder whose disturbance exceeds the threshold can send a correction signal to the other cylinders. Each cylinder independently accumulates the correction amount from other cylinders, adapting to the situation where multiple cylinders exceed the threshold simultaneously in complex off-center load scenarios.

[0050] S2.7, Estimation layer output: Each hydraulic cylinder operates in parallel using AEKF-UI, outputting: (1) Estimated value of fusion state (displacement) ,speed Hydraulic pressure ); (2) Estimated value of unknown disturbance force ; (3) Posterior covariance matrix and the health index calculated from it (Equation 13); (4) Standard deviation of displacement estimation (The square root of the diagonal elements of the corresponding displacement components in the covariance matrix) serves as a confidence quantification index.

[0051] S3: Using the multidimensional fusion feature vector as input, an anomaly score is performed using an improved isolated forest based on a piecewise cumulative probability density function, and the anomaly score and its corresponding warning level are output. like Figure 4 The diagram shown is a flowchart of the PCPD-iForest anomaly detection process provided in this embodiment of the invention. The improved isolated forest anomaly detection method (PCPD-iForest) based on the piecewise cumulative probability density function is designed as follows: S3.1, Mechanism for generating segmentation values ​​based on power-weighted intervals: The improved method for generating cut values ​​in isolated forests based on the piecewise cumulative probability density function is as follows: When splitting at each node of the isolated tree, the probability of selecting each interval is determined along the selected feature axis according to a power function of the length of each data interval, ensuring that the cut values ​​preferentially fall in the sparse gaps between data points. After determining the selected intervals, cut values ​​are randomly generated within those intervals using a uniform distribution. The specific process is as follows: When splitting at each node of the isolated tree, the current node is processed along the selected feature axis. Sort unique value sequence , No. segment interval The probability of being selected is: (16) in, The feature axis selected for splitting the current node; The current node along the selected feature axis A sorted sequence of unique values; The sorted sequence of the first... A unique value; The sorted sequence of the first... A unique value; : No. segment interval The probability of being selected as the interval containing the cutoff value; : The power exponent of the interval length , used to control the strength of gap preference, recommended value range [1,3]; : Summation of index variables, iterating from 1 to The interval numbering; The current node along the selected feature axis The total number of unique values ​​in the sorted sequence, and the number of intervals are ; : No. The length of the segment interval.

[0052] Equation (16) controls which interval the cut falls into (gap preference). Within the determined interval... Inside, cutting value Generated using a uniform distribution: (17) in, The right endpoint of the interval has been determined (the first...) (a unique value); The left endpoint of the interval has been determined (the first...) (a unique value); : The final generated cut value; :obey Uniformly distributed random numbers.

[0053] Equations (16) and (17) have separate responsibilities: the former realizes the interval selection of gap preference, while the latter performs uniform sampling within the selected interval without iterative solution.

[0054] Engineering significance: Under normal construction conditions, the combined pressure-displacement characteristic data of each cylinder are highly aggregated, forming multiple dense clusters. Equation (16) uses a power function of interval length for weighting, ensuring that the cut-off values ​​preferentially fall into the sparse gaps between clusters, thus allowing progressive drift anomalies to be isolated with shorter paths and obtaining higher anomaly scores. The larger the gap, the stronger the preference; recommended value range The cluster density distribution characteristics of the training set data can be adjusted using the validation set.

[0055] S3.2, U-shaped probability density compensation mechanism: The U-shaped probability density compensation mechanism in step S3.2 is as follows: During the generation of random numbers for node splitting, a U-shaped probability density correction is superimposed to increase the probability that the cut value falls in the end region of the feature space, thereby enhancing the detection sensitivity for safety boundary anomalies such as pressure exceeding limits or travel exceeding limits. The specific process is as follows: To address safety boundary anomalies at the ends of characteristic distributions such as excessive cylinder pressure and stroke exceeding limits, a U-shaped probability density correction is superimposed during the node splitting random number generation step. (18) in, The random number generated for splitting the current node (scaled and mapped to the actual feature interval) : U-shaped probability density function, used to superimpose corrections in the node splitting random number generation step, so that the cutting probability of the end region is relatively increased; : Control coefficient for U-shaped probability density correction, range of values The recommended default value is 0.5, which can be adjusted by maximizing the end-point anomaly F1 score on the debugging data. It degenerates into a uniform distribution (the case of a standard isolated forest). For the standard normal kernel function: (19) The standard deviation of the normal kernel is calculated dynamically node by node by taking the standard deviation of the training data along the current feature axis.

[0056] Equation (18) makes exist The value is higher at the (end) point. The lower value at the center increases the probability that the cut will fall in the end region, thus improving the detection sensitivity of safety boundary anomalies.

[0057] S3.3, Deviation Enhancement Anomaly Scoring Function (DEAS): In step S3.3, the bias-enhanced anomaly scoring function is used to calculate the final anomaly score. This function includes both path score and bias score. The path score reflects the ease with which a sample is isolated, while the bias score reflects the degree to which the sample's feature values ​​deviate from the normalized average of the splitting thresholds of each node. The final anomaly score is the product of the path score and the bias score. The specific process is as follows: Let the sample In an isolated tree The traversal path in the middle has experienced a total of The split node, the first The splitting threshold at each node is: The value range of this node's data pool on the corresponding feature axis is... (For normalization), the projection value of the sample onto this feature axis is The deviation scoring item is defined as follows: (20) in, : Sample data points to be scored; : Index number of the isolated tree; :sample In the The total number of split nodes encountered during the traversal of a path in an isolated tree; : The first in the traversal path Index of each split node; : No. The splitting threshold at each splitting node; : No. The range of values ​​for each split node's data pool on the corresponding feature axis is equal to... ; : No. The maximum value of the data pool of each split node on the corresponding feature axis; : No. The minimum value of the data pool of each split node on the corresponding feature axis; :sample In the The first isolated tree Each split node corresponds to a projection value on the feature axis; :sample In the The bias score term on the isolated tree is a dimensionless quantity, ranging from [0,1], representing the degree to which the sample feature values ​​deviate from the normalized average of the splitting thresholds of each node. Comprehensive scoring function: (twenty one) in, :sample In the isolated forest The overall abnormality score; An isolated forest is a collection of many isolated trees. An operator that takes the expectation (average) of all isolated trees; :sample In the The path length in an isolated tree, i.e., the number of node splits traversed; : The normalization factor for path length ; Harmonic function ; The subsampling size for isolated forests is consistent with that of standard isolated forests. Path score reflects the difficulty of isolation; range of values. The shorter the path, the easier it is to be isolated, and the higher the score. : Deviation points, that is value range This reflects the extent to which the characteristic value deviates from the normal distribution.

[0058] The design rationale for the product form is that the product of path score and deviation score ensures that only samples that simultaneously meet the criteria of being "easily isolated" (short path) and "significantly deviating" (large deviation) can obtain a high comprehensive score. This effectively suppresses normal marginal samples that accidentally obtain high path scores due to data sparsity—although these samples have short paths, their feature values ​​do not significantly deviate from the node splitting threshold, resulting in a small deviation score and a low comprehensive score after product. This dual constraint mechanism is particularly effective against progressive drift anomalies.

[0059] Typical value range: During the commissioning period verification, under normal operating conditions... The mean is approximately 0.1~0.2 (the eigenvalues ​​are concentrated near the splitting threshold of each node), and abnormal operating conditions... The mean is approximately 0.4 to 0.7 (the eigenvalues ​​deviate significantly from the normal distribution range). Overall score. Distribution statistics (mean) Standard deviation (This is obtained during debugging and updated regularly.)

[0060] S3.4 Definition of Multi-Level Early Warning Thresholds: The mean of DEAS scores for each sample under normal operating conditions during the commissioning period was statistically analyzed. and standard deviation The three-level early warning thresholds are defined as shown in Table 1: Table 1 Threshold definitions based on standard deviation multiples have good adaptability across different working conditions—the same threshold system can be reused for the same equipment at different construction stages (changes in foundation conditions, beam weight); however, it is recommended to recalculate the thresholds after replacing critical components (such as hydraulic seals) or after significant changes in working conditions. and To maintain effectiveness.

[0061] S3.5, Detection Layer (PCPD-iForest): The multi-dimensional fusion feature vector output by AEKF-UI (including estimated displacement, estimated pressure, estimated value of unknown disturbance force, and health index of each cylinder) is sent to PCPD-iForest for DEAS scoring, and the warning is triggered according to the three-level threshold classification of S3.4.

[0062] S3.6 Training Data Construction and Parameter Configuration: Training samples are collected from normal operating condition data during the equipment commissioning phase and early construction cycles. The sampling frequency is set according to the sensor type (1 Hz is recommended for low-speed parameters such as pressure and displacement, and 10 Hz is recommended for high-speed parameters such as attitude and acceleration). Abnormal samples are constructed by artificially simulating typical faults, including: gradual drift of pressure sensors, jamming of hydraulic cylinder travel tracks, and slow pressure drop due to seal leakage. The training set should ideally cover normal operating condition data from at least several consecutive complete construction cycles. A minimum of 50 isolated trees and 256 subsamples are recommended, with depth restrictions. .

[0063] S4: Based on the estimated value of the unknown disturbance force, perform multi-cylinder load balancing adjustment and PID synchronous control to generate target control commands for each cylinder in order to correct the multi-cylinder synchronization error; Step S4 is used to implement the control closed loop.

[0064] S4.1 Multi-cylinder load balancing adjustment: In step S4.1, the multi-cylinder load balancing adjustment specifically includes: Assume the estimated value of the unknown disturbance force for each cylinder is... ,calculate Arithmetic mean of the estimated unknown disturbance forces of all cylinders at any given time : (twenty two) in, : No. The cylinder is The estimated value of the unknown disturbance force at time; Total number of hydraulic cylinders; : The arithmetic mean of the estimated unknown disturbance forces of all cylinders at any given time; According to the arithmetic mean With the The cylinder is Estimated value of unknown disturbance force at time The deviation is calculated as follows: Cylinder pressure correction : (twenty three) in, : No. The cylinder is Pressure correction amount at any time; Load balancing gain, recommended value range is 0.05~0.15, tuned according to the synchronous convergence speed (too large a gain can easily cause overshoot, too small a gain will slow convergence). : No. The piston area of ​​the cylinder is determined by the equipment specifications.

[0065] The pressure correction amount Superimposed on the system reference pressure command Above, generate the first The cylinder's ultimate target pressure : (twenty four) in, : No. The cylinder is The pressure of achieving the ultimate goal at any given moment; : The system reference pressure command at any given time.

[0066] Ultimate goal pressure The alarm limit must not be exceeded as specified in the equipment specifications (usually 1.2 times the rated pressure). If it is exceeded, the alarm will be cut off to the upper limit and a secondary warning will be triggered simultaneously.

[0067] S4.2, PID Synchronous Control and Attitude Compensation: Using the actual displacement of each cylinder estimated by AEKF-UI as the feedback quantity, when the synchronization error of multiple cylinders exceeds the tolerance range (refer to engineering specifications, the synchronization error of the lifting master cylinder is not greater than 1 mm), the PID controller automatically triggers the differential compensation command. The travel correction uses the fused lateral deviation estimate as input, and the lateral and longitudinal deviation control targets meet the requirements of engineering specifications (reference values: lateral not greater than 2 mm, longitudinal not greater than 5 mm).

[0068] S5: The fusion state estimate, unknown disturbance force estimate, displacement estimate standard deviation, health index, anomaly score and early warning level are mapped to the digital twin platform in real time through a standardized data protocol, driving the three-dimensional twin model to perform state synchronization and visual monitoring.

[0069] Step S5 is used to receive the state estimate, confidence level, anomaly score and warning level of each cylinder in a standardized data packet through the digital twin platform, drive the displacement animation of each cylinder component in the WebGL-based 3D twin, map the warning level to the component color, and support historical frame playback and full process data traceability.

[0070] The data protocol for the digital twin platform is as follows: The digital twin platform and the physical equipment achieve real-time bidirectional mapping through standardized data packets, and the MQTT (QoS=2) protocol ensures reliable message transmission. An example of the data packet structure for each hydraulic cylinder is as follows: { "deviceID":"Crane_001", "timestamp":"2024-03-15T10:30:45Z", "cylinders":[ { "id":"Cyl_1", "est_displacement_mm":45.23, "std_displacement_mm":0.12, "est_pressure_MPa":24.51, "disturbance_force_kN":12.3, "health_index":0.93, "anomaly_score": 0.42, "alarm_level":0 } ] } Field descriptions: std_displacement_mm is obtained by taking the square root of the diagonal elements of the displacement components in the AEKF-UI posterior covariance matrix; anomaly_score is the DEAS score; alarm_level takes values ​​of 0 / 1 / 2 / 3, corresponding to normal / level 1 / level 2 / level 3 warning respectively; health_index is calculated by equation (13); disturbance_force_kN is the estimated value of the unknown disturbance force. Data packets are sent at fixed intervals, and the sending frequency is set according to the sensor sampling rate and control real-time requirements.

[0071] Example 2: (I) Application of AEKF-UI in synchronous control of 8-channel lifting cylinders: G16090-120 type lifting cylinder (stroke 120mm, rated pressure) For example, (8-cylinder linkage).

[0072] The joint state vector includes: cylinder displacement ,speed Hydraulic chamber pressure Chebyshev polynomial coefficients (Third order, input displacement normalized to Equation 11) Substitutes and unknown disturbances There are 7 dimensions in total. Initial parameters are set according to the nominal values ​​under rated operating conditions, and allowable... Initial estimation error; Calculate in real time according to formula (10), (After the debugging phase is completed, a one-time adjustment can be performed based on the measurement residual statistics); The matrix is ​​calibrated by single-cylinder loading test according to formula (16) during the factory commissioning stage.

[0073] During the equipment commissioning phase, the AEKF-UI underwent five complete lifting cycles of verification. After approximately 3 cycles, the parameters converged, and the standard deviation of the stroke synchronization error of the 8-cylinder system converged to the range of 0.2~0.5mm (meeting the specification requirement of ≤1mm), with the pressure difference between each cylinder controlled within 0.3 MPa. The above data are the measured range during the commissioning period. The actual accuracy during formal construction is affected by factors such as the rate of load change and foundation conditions, and the results of on-site calibration should be taken as the standard.

[0074] The displacement sensor has a rated accuracy of 0.5 mm (range). ~ The pressure sensor has a rated accuracy of 0.1 MPa (range). ~ The above refers to the rated accuracy of the sensor. The multi-cylinder synchronization accuracy that the control system can achieve depends on the construction environment. The sensor's rated accuracy should not be equated with the system's control accuracy.

[0075] (II) Application of PCPD-iForest in abnormal detection of walking systems: This invention's method is not only applicable to the synchronous control of lifting cylinders, but can also be extended to the anomaly detection of traveling cylinders. Taking the G14080-600 type traveling cylinder (stroke 600 mm, lateral deviation tolerance ±0.5 mm, longitudinal movement speed approximately 50 mm / min) as an example, a training set is constructed using the multi-dimensional state features (lateral deviation estimate, longitudinal velocity estimate, track pressure estimate, and disturbance force estimate) fused with AEKF-UI, and a PCPD-iForest model is established. , An isolated tree, with a subsample size of 256. Normal kernel (Dynamic calculation per node).

[0076] The training set includes normal samples from several consecutive construction conditions; abnormal samples are constructed by artificially simulating three typical faults: track jamming, sensor drift, and lateral deviation exceeding limits. In the debugging and validation set containing these three types of anomalies, PCPD-iForest improves the detection rate of progressive anomalies within dense construction conditions (by approximately 5-10 percentage points) compared to the standard isolated forest, and also enhances the sensitivity to end-point exceeding-limit anomalies; the algorithm training time increment is typically less than twice, meeting the real-time requirements of edge computing nodes. The above detection rate improvement data are the average results of the three types of simulated faults in the debugging and validation; the specific values ​​depend on the magnitude and rate of fault injection. In formal construction, the actual labeled data should be used for verification; the warning threshold should be recalculated before formal construction. and .

[0077] (III) Integration of Digital Twin Platform: The digital twin platform receives the combined output of each subsystem using standardized data packets defined in 3.5 via the MQTT QoS=2 protocol.

[0078] 3D Twin Modeling and Visualization: The 3D twin model is exported from the beam-building machine CAD design file (containing the 3D geometric data of the traveling mechanism, lifting cylinders, side molds, bottom basket, and rear anchor mechanisms) to a standard format and loaded into a 3D engine supporting WebGL (such as a web client based on Three.js or IFC.js); the extension of each cylinder is driven in real-time by the displacement estimated by AEKF-UI, which drives the displacement animation of the cylinder components, with a refresh rate of no less than 1 Hz; the alarm_level field of the anomaly score is mapped to the component color (…). green, yellow, orange color, (Red) Supports historical frame playback, allowing operators to intuitively assess the health status of each system. Historical data supports full-process traceability, and the convergence curve of the AEKF-UI polynomial coefficients can be used to analyze the long-term trend of cylinder nonlinear characteristics, providing a reference for preventative maintenance. For example... Figure 5 The image shown is the rear anchor lifting outrigger lifting control page (including a 3D model schematic diagram) of Embodiment 2 of the present invention.

[0079] In summary, the method provided by the embodiments of the present invention effectively improves the state perception capability, anomaly identification capability, and control coordination during the construction process of a cantilever beam-making machine through a unified closed loop of state estimation, anomaly detection, load balancing control, and digital twin monitoring.

[0080] Example 3: Figure 6This is a schematic diagram of the structure of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, and may also represent various forms of mobile devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the invention described and / or claimed herein.

[0081] like Figure 6 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as ROM 12 or RAM 13, communicatively connected to the at least one processor 11. ROM is a read-only memory, and RAM is a random access memory. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in ROM 12 or loaded into RAM 13 from storage unit 18. RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via bus 14. I / O interface 15 is also connected to bus 14.

[0082] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0083] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the methods described above.

[0084] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0085] The above embodiments are merely illustrative examples and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for multi-source data fusion control and intelligent monitoring of cantilever beam-making machines based on digital twins, characterized in that, include: S1: Real-time acquisition of displacement measurement values, pressure measurement values, and construction vibration intensity parameters of each cylinder of the cantilever beam building machine; S2: For each hydraulic cylinder, using the displacement and pressure measurements as measurement inputs, a recursive filtering process including nonlinear state prediction and measurement updates is executed. The attenuation memory weights of the filter are dynamically adjusted using the construction vibration intensity parameters, and an improved extended Kalman filter with attenuation memory and unknown input decoupling estimation is used to perform joint state estimation of the cylinder's displacement, velocity, pressure, and unknown disturbance force. The output includes a multi-dimensional fusion feature vector containing the fused state estimate, the unknown disturbance force estimate, the health index calculated based on the covariance matrix, and the standard deviation of the displacement estimate. The nonlinear state prediction specifically includes: predicting the joint state vector of the hydraulic cylinder using a nonlinear state transition function, which characterizes the hydraulic cylinder dynamics including nonlinear friction force; the attenuated memory improved extended Kalman filter with unknown input decoupling estimation specifically includes: after completing the nonlinear state prediction at each time step, using the measurement residual to calculate the unknown disturbance force estimate in real time using the least squares method, realizing online decoupling between the unknown disturbance force and the system state, the calculation formula is: ； in, The unknown perturbation force at time k+1 The estimated value, The unknown input influence matrix is ​​determined by the force geometry of the hydraulic cylinder. This is the observation vector containing displacement and pressure measurements at time k+1. For nonlinear observation functions, The state vector at time k+1 The prior estimate; The joint state estimation also employs Chebyshev polynomials to model the nonlinear frictional force of the hydraulic cylinder online, specifically including: Displace the hydraulic cylinder Normalize to interval [ After [1,1], the normalized cylinder displacement is obtained. Then expressed as a weighted sum of Chebyshev polynomials: ； in, for Chebyshev polynomials The coefficients to be identified are... For the nonlinear frictional force of the hydraulic cylinder, Let be the order of the Chebyshev polynomial; S3: Using the multidimensional fusion feature vector as input, an anomaly score is performed using an improved isolated forest based on a piecewise cumulative probability density function, and the anomaly score and its corresponding warning level are output. S4: Based on the estimated value of the unknown disturbance force, perform multi-cylinder load balancing adjustment and PID synchronous control to generate target control commands for each cylinder in order to correct the multi-cylinder synchronization error; S5: The fusion state estimate, unknown disturbance force estimate, displacement estimate standard deviation, health index, anomaly score and early warning level are mapped to the digital twin platform in real time through a standardized data protocol, driving the three-dimensional twin model to perform state synchronization and visual monitoring.

2. The method according to claim 1, characterized in that, In step S2, the attenuation memory weight of the filter is dynamically adjusted using the construction vibration intensity parameter, employing an attenuation memory weight mechanism, specifically including: Based on the construction vibration intensity parameters at the current moment Vibration intensity compared to normal construction reference The ratio of the decay memory weight coefficient is dynamically calculated. and the decay memory weight coefficient In the recursive process of introducing the prediction error covariance matrix, the influence weight of historical measurement data on the current state estimation is reduced when vibration intensifies. The formula is as follows: ； in, To estimate the error covariance matrix a priori, Let Jacobian be the state transition matrix. For the posterior estimation of the error covariance matrix, Let be the process noise covariance matrix.

3. The method according to claim 1, characterized in that, Step S2 also includes a multi-cylinder coupled disturbance broadcast compensation mechanism, specifically including: When the The estimated value of the unknown disturbance force of the cylinder satisfies The broadcast is triggered at a certain time, in which, The preset threshold coefficient, The rated reference disturbance force; After the broadcast is triggered, the estimated value of the unknown disturbance force of the j-th cylinder is corrected to obtain the value of the j-th cylinder. The cylinder is Estimated value of unknown disturbance force at time , used for the Correction of cylinder state prediction for the next moment: ； in, The off-center load coupling coefficient was pre-calibrated through single-cylinder loading tests during the commissioning phase, and .

4. The method according to claim 1, characterized in that, In step S3, the improved isolated forest cut value generation method based on the piecewise cumulative probability density function includes: When splitting at each node of the isolated tree, the probability of each interval being selected is determined along the selected feature axis according to the power function of the length of each data interval, so that the cut value falls preferentially in the sparse gap region of the data points; after determining the selected interval, the cut value is randomly generated in a uniform distribution within the selected interval.

5. The method according to claim 1, characterized in that, In step S3, the improved isolated forest based on the piecewise cumulative probability density function also includes a U-shaped probability density compensation mechanism: During the generation of random numbers for node splitting, a U-shaped probability density correction is superimposed to increase the probability that the cut value falls in the end region of the feature space, thereby enhancing the detection sensitivity for pressure over-limit or travel over-limit anomalies.

6. The method according to claim 1, characterized in that, In step S3, the improved isolated forest based on the piecewise cumulative probability density function uses the bias-enhanced anomaly scoring function to calculate the final anomaly score. The deviation enhancement anomaly scoring function includes both path score and deviation score. The path score reflects the ease with which a sample is isolated, and the deviation score reflects the degree to which the sample feature value deviates from the normalized average of the splitting threshold of each node. The final anomaly score is the product of the path score and the deviation score.

7. The method according to claim 1, characterized in that, The health index is defined as: the mean of the steady-state covariance matrix of the i-th cylinder under normal operating conditions during the commissioning period. traces and current Posterior covariance matrix at time 1 The ratio of traces: ； in, For the first The cylinder is Health index at any time Let be the trace of the matrix.

8. The method according to claim 1, characterized in that, In step S4, the multi-cylinder load balancing adjustment specifically includes: calculate Arithmetic mean of the estimated unknown disturbance forces of all cylinders at any given time ; According to the arithmetic mean With the The cylinder is Estimated value of unknown disturbance force at time The deviation is calculated as follows: Cylinder pressure correction: ;in, For load balancing gain, The piston area; The pressure correction amount Superimposed on the system reference pressure command Above, generate the first The cylinder's ultimate target pressure ; The final target pressure shall not exceed the alarm limit specified in the equipment specification. If it exceeds the limit, the alarm will be truncated to the alarm limit and a level 2 warning will be triggered.

Images