Intelligent pile hammering energy optimization control method and system based on machine learning

Abstract

This invention discloses a machine learning-based intelligent pile driver hammering energy optimization control method, comprising the following steps: after each hammering, collecting relevant data and constructing a time-series feature vector; inputting the time-series feature vector into a prediction model to predict the total hammering energy; and based on the predicted total energy, obtaining the single-pile-driving energy through a pile-driving strategy for pile driving. This invention, through data-driven prediction and optimization, automatically finds and executes the lowest energy point that meets pile-driving requirements, significantly reducing fuel / electricity consumption, avoiding fluctuations and lags caused by manual operation, and ensuring that the hammering process remains within the scientifically optimal range, thus improving the stability of pile-driving efficiency and pile quality. Through transfer learning, the system can quickly adapt to new and complex geological conditions using historical experience and a small amount of new data, greatly enhancing the system's generalization ability and engineering practicality.

Description

Technical Field

[0001] This invention belongs to the field of automation and intelligent control technology of construction machinery, specifically relating to a machine learning-based intelligent pile driver hammer impact energy optimization control method and system. Background Technology

[0002] Hydraulic pile hammers are the core equipment in pile foundation construction, and precise control of their hammering energy is crucial for ensuring construction quality, efficiency, and energy conservation. Currently, hammering energy control heavily relies on the operator's personal experience, resulting in the following inherent defects: High subjectivity and lack of quantitative basis: Operators adjust energy based on vague information such as pile rebound and sound, failing to achieve scientific quantitative matching. High energy consumption and unstable efficiency: To avoid pile driving failure, a conservative "better too much than too little" strategy is often adopted, leading to energy waste; insufficient energy requires multiple hammer blows, reducing efficiency. Poor adaptability: Faced with complex and changing strata (such as transitioning from loose sand to hard clay), manual response is lagging, making it difficult to find the optimal energy point in real time, easily causing pile damage or failure to reach the designed depth. Current monitoring of pile foundation construction mostly remains at the data recording or post-event analysis stage, lacking a closed-loop intelligent control system capable of real-time prediction of pile driving response, dynamic optimization of current hammering energy, and rapid adaptation to new construction environments.

[0003] Existing technologies have included some research attempts to improve pile driver control through automation or semi-automation. For example, the hydraulic pile driver control method and system (CN202011499068.0) developed by Taiyuan Heavy Machinery (Tianjin) Binhai Heavy Machinery Co., Ltd. automatically matches the control parameters of the hydraulic and drive devices through modular energy and continuous driving frequency inputs, achieving simplified operation and a certain degree of power matching. However, this method still relies on preset modular empirical rules and does not introduce a data-driven optimization mechanism. It cannot dynamically adjust the hammering energy according to real-time construction status and geological conditions, and remains within the scope of static parameter matching. Therefore, the pile driving strategy may change in the face of changes in the construction environment (such as soil layers, soil hardness, and depth).

[0004] On the other hand, a method for matching control parameters of hydraulic pile hammer construction based on simulation data and small-scale actual machine samples (Research on Parameter Matching Method of Hydraulic Pile Hammer Control System, Guo Nengchang, South China University of Technology) uses a surrogate model and transfer learning algorithm, which achieves parameter prediction and optimization to a certain extent. However, this research mainly relies on hydraulic system simulation data and lacks a complete model of the "hammer-pile-soil" coupled system. Furthermore, there are inherent deviations between the simulation model and the actual machine data, which limits its generalization ability and prediction accuracy under actual complex geological conditions.

[0005] In summary, the existing technologies still have the following shortcomings: First, they lack dynamic, data-driven modeling methods for the hammer-pile-soil system; second, they rely too much on simulation or empirical rules, making it difficult to achieve real-time, adaptive, and high-precision hammer energy optimization control; and third, they have not formed a complete intelligent control closed loop from data acquisition and predictive modeling to real-time optimization.

[0006] Therefore, there is an urgent need for an intelligent control method that can automatically learn, predict and optimize hammering energy in real time based on actual machine data, in order to solve the above-mentioned shortcomings and improve the level of intelligence and energy efficiency of pile foundation construction. Summary of the Invention

[0007] This invention aims to solve the technical problems of high energy consumption, low efficiency, unstable quality, and inability to adapt to complex geological conditions caused by the reliance on manual experience in the control of existing hydraulic pile hammers. It provides a data-driven intelligent control method and system that can automatically, in real-time, and accurately match the optimal hammering energy.

[0008] A machine learning-based method for optimizing and controlling the hammer impact energy of a smart pile driver includes the following steps:

[0009] Step 1: After each hammer blow, collect the corresponding data and construct a time-series feature vector: Step 2: Input the temporal feature vector into the prediction model to predict the total energy of the hammer impact; Step 3: Adopt a transfer learning mechanism for online adaptation to address the cold start problem of data at new construction sites; Step 4: Based on the predicted total energy, obtain the single pile driving energy through the pile driving strategy and drive the pile; Step 5: Based on the SHAP algorithm, perform interpretability analysis on the prediction model and output the feature importance ranking and decision basis.

[0010] Furthermore, the collected data includes: Operating parameters: current pile penetration depth, hammer type parameters, pile diameter, and wall thickness; Energy sequence: current and previous history Energy of each hammer blow Response sequence: The penetration sequence corresponding to the historical hammer blows; Geological characteristics: soil layer type and hardness.

[0011] Furthermore, the prediction model employs long short-term memory networks, Transformers, or ensemble tree models.

[0012] Furthermore, the pile driving strategy in step 3 is to calculate the predicted total energy value by using different single hammer blow energies under the current working conditions within the single safe hammer blow energy. The single hammer blow energy corresponding to the minimum total energy value is taken as the pile driving strategy, and pile driving is carried out using this single pile driving energy.

[0013] Furthermore, the prediction model uses historical construction site data during the training phase and employs mean squared error or mean absolute error as the loss function for supervised learning.

[0014] Furthermore, the prediction model employs a transfer learning mechanism for online adaptation to address the cold start problem of new construction site data.

[0015] Furthermore, a transfer learning mechanism is employed: Pre-training phase: Train a general base model using pile driving data from multiple historical construction sites; Fine-tuning phase: Before construction begins at the new site, test pile data is collected as target domain data to fine-tune the foundation model.

[0016] The system for implementing the intelligent pile driver hammer impact energy optimization control method based on machine learning includes: a sensing unit and a control unit; The sensing unit is used to collect vibration signals during the hammering process, measure the pile penetration depth in real time, monitor the status of the hydraulic system and calculate the actual hammering energy, and collect environmental and working conditions. The control unit, using a PLC or industrial computer, is responsible for data acquisition, feature extraction, model prediction and optimization calculation, and receives energy commands from the controller to adjust the hammering energy of the hydraulic hammer in real time.

[0017] A computer device according to the present invention includes a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, which, when executed by the processor, causes the processor to implement the method described herein.

[0018] The present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor implements the method described herein.

[0019] Compared with existing technologies, the beneficial effects of the present invention are as follows: Precise energy saving: Through data-driven prediction and optimization, it automatically finds and executes the minimum energy point that meets the pile driving requirements, significantly reducing fuel / electricity consumption.

[0020] Improved quality and efficiency: It avoids the fluctuations and delays of manual operation, keeps the hammering process in the scientifically optimal range, and improves the stability of pile driving efficiency and pile quality.

[0021] Intelligent Adaptation: Through transfer learning, the system can quickly adapt to new and complex geological conditions by utilizing historical experience and a small amount of new data, which greatly improves the system's generalization ability and engineering applicability.

[0022] Reduced reliance on experience: This reduces the dependence on highly skilled workers in construction, providing core technical support for achieving unmanned and standardized pile foundation construction. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a flowchart illustrating a machine learning-based intelligent pile driver hammer impact energy optimization control method as an example.

[0025] Figure 2 The flowchart illustrates the optimization process of the prediction model in this example.

[0026] Figure 3 The energy prediction model test cluster map is shown in the example.

[0027] Figure 4 The scatter plot before and after migration is shown in the example. Detailed Implementation

[0028] To enable those skilled in the art to better understand the present invention, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0029] This embodiment of a machine learning-based intelligent pile driver hammering energy optimization control system includes: The sensing unit includes a triaxial accelerometer (model such as PCB 356A01) mounted symmetrically on the side of the hammer and the top of the pile. The sampling frequency is no less than 10 kHz. Shock-resistant encapsulation and temperature compensation mechanisms ensure signal stability under strong vibration, high humidity, and dusty environments. Vibration signals are hardware filtered (500Hz low-pass) and transmitted to the control unit via the CAN bus for inverting the hammer-pile contact state and energy transfer efficiency. A high-precision laser rangefinder (such as SICK DT1000) is mounted on the pile frame beam, vertically aligned with the pile top mark, to measure the pile top elevation in real time and calculate the penetration depth based on the pile length. The system is equipped with automatic calibration and temperature compensation algorithms, achieving a measurement accuracy of ±0.5 mm, and data is collected once per hammering cycle. Hydraulic system sensors, including pressure and flow sensors, monitor the hydraulic system status and calculate the actual hammering energy.

[0030] Environmental and working condition sensors: Optional soil resistance sensors, pile attitude sensors, etc.

[0031] Control unit, including: Industrial controller: Using PLC or industrial computer, it is responsible for data acquisition, feature extraction, model prediction and optimization calculation.

[0032] Electro-hydraulic proportional control valve: Receives energy commands from the controller and adjusts the hammering energy of the hydraulic hammer in real time. Human-machine interface: Used for parameter setting, status monitoring, and alarm prompts.

[0033] like Figures 1-2 As shown in the figure, the intelligent pile driver hammering energy optimization control method based on machine learning in this embodiment specifically includes the following steps: Step 1: Extract multi-dimensional features in real time and construct a time-series feature vector sequence.

[0034] After each hammering, the following data is collected in real time, and a temporal feature vector is constructed.

[0035] Operating parameters: current pile penetration depth Y, hammer type parameter M, pile diameter D, and wall thickness H.

[0036] Energy sequence: current and previous history Next (such as) Hammering energy value ; Response sequence: Penetration sequence corresponding to historical hammer blows ; Geological characteristics: soil layer type, hardness, etc.

[0037] Preprocessing: Outliers are filtered out, and all features are standardized to form a time-series feature vector. This is used in subsequent prediction models. The time-series feature vector... It includes not only the current hammering parameters, energy sequence, response sequence, and geological features. In one embodiment, a sliding window mechanism is used to introduce forward... =5 hammer historical states, forming a feature sequence of length n: ,in eigenvectors at time step This includes: continuous features, such as depth. ,energy Discrete coding features: such as soil layer type (One-hot coding), hammer type number, pile diameter category; Statistically derived features: such as the average energy of the past 3 hammer blows, penetration variance, and depth variation trend (linear fitting slope).

[0038] Step 2: Use a predictive model to predict the hammer impact response.

[0039] Time series feature vectors Input into a pre-trained machine learning model to predict the energy of the current single burst. Below, the pile body from the current depth Sink to target depth Total energy required for each stage .

[0040] In one embodiment, the machine learning model is selected from models that can handle temporal dependencies, such as Long Short-Term Memory Networks (LSTM), Transformers, or ensemble tree models (such as XGBoost).

[0041] Model inputs include at least the current hammering energy, historical energy sequence, penetration sequence, current depth, and geological features.

[0042] Model output: Total energy during the prediction phase ,in This represents the generalization function of the selected model.

[0043] The machine learning models use historical construction site data during the training phase, employing mean squared error (MSE) or mean absolute error (MAE) as the loss function for supervised learning. Table 1 compares the prediction performance of each algorithm based on the test set data. Figure 3Scatter plots of the algorithms were generated (the closer the predicted values ​​are to the diagonal, the better the performance). The MAE (mean absolute error), MSE (mean squared error), and R² (coefficient of determination) of four models—LSTM, Transformer, XGBoost (XGB), and Random Forest (RF)—were compared on the test set. XGBoost showed the best performance in MAE (195.6 kJ) and MSE (146,351 kJ²).

[0044] Step 3: Employing transfer learning mechanisms for rapid online adaptation to address the "data cold start" problem at new construction sites. The transfer learning mechanism includes: Using the prediction model trained in step 2 as the base model, Table 1 shows that the XGB model and the RF model perform better. Therefore, the XGB model and the RF model are used as the base models for this transfer learning. Before construction at the new site, a small amount of test pile data was collected as the target domain data to fine-tune the base model: using the pre-trained model as the initial state, training continued on the new data. Rapid adaptation: A model Ftransfer(x) adapted to the new geological conditions can be obtained with only a small amount of data, significantly improving prediction accuracy and generalization ability. Table 2 shows a comparison of the performance of the two models before and after transfer learning on the new target domain data. Figure 4 The table shows scatter plots of the predicted points before and after migration for the two algorithms. The horizontal axis represents the actual total energy, and the vertical axis represents the predicted total energy. (a) shows the XGB model before fine-tuning, (b) shows the XGB model after fine-tuning, (c) shows the RF model before fine-tuning, and (d) shows the RF model after fine-tuning. After fine-tuning, the predicted points are closer to the diagonal, and the dispersion is significantly reduced. (Table 2 and...) Figure 4 XGB performed best. After fine-tuning, the MAE of the XGB model decreased from 695.58 to 195.56, and the relative error decreased from 24.99% to 8.08%, showing the most significant improvement and the highest accuracy. Therefore, XGB was selected as the energy prediction model.

[0045] Table 1 Comparison of Energy Prediction Results

[0046] Table 2. Results of Transfer Learning Algorithms

[0047] Step 4: Optimize the rolling time domain energy based on the prediction results.

[0048] Based on the prediction model, a real-time optimization model is established to optimize the problem: within the safe single hammer energy, the predicted total energy value is solved by using different single hammer energies under the current working conditions. The single hammer energy corresponding to the minimum total energy value is used as the pile driving strategy, and pile driving is carried out using this single pile driving energy.

[0049] The objective function of the real-time optimization model is to minimize energy consumption (or total energy consumption), with the constraint of ensuring that the predicted penetration is within a preset safe and efficient range. A grid search strategy is adopted to solve for the optimal energy sequence for the next finite number of steps in each control cycle, and only the optimal hammering energy command is issued to the actuator.

[0050] Step 5: Model Interpretability Analysis Based on SHAP Algorithm. To enhance the credibility and acceptability of the model in engineering practice, this system introduces the SHAP (Shapley Additive Explanations) algorithm to interpret the prediction model. SHAP, based on game theory, can quantify the contribution of each input feature to the model's prediction results, thus providing engineers with intuitive decision-making support. The specific implementation is as follows: After each hammer energy prediction, the system calls the SHAP interpreter to parse the current time-series feature vector Xt; calculates the SHAP value of each feature (such as soil type, historical penetration, current depth, etc.) and generates a feature importance ranking; displays the feature contribution map through a visual interface to help construction personnel understand why the model recommends a certain hammer energy; when the model prediction is abnormal, SHAP analysis can quickly locate possible causes (such as abnormal sensor data or geological changes).

[0051] Taking the construction of PHC pipe piles at a certain construction site as an example: The system is initialized by loading a pre-trained model (trained based on data from five previous construction sites). Three test piles are driven at the new site, collecting approximately 200 sets of hammer impact data. The LSTM model is fine-tuned using the test data (the last four layers are unfrozen, and the learning rate is 0.0001). During formal construction, the system collects vibration, depth, and pressure data in real time. Steps S1-S3 are executed after each hammer impact, dynamically outputting the optimal hammer impact energy. Data is continuously collected during construction, and the model is incrementally updated every 10 piles completed to further improve adaptability and accuracy.

[0052] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A machine learning-based intelligent pile driver hammer impact energy optimization control method, characterized in that, Includes the following steps: Step 1: After each hammer blow, collect the corresponding data and construct a time-series feature vector: Step 2: Input the temporal feature vector into the prediction model to predict the total energy of the hammer impact; Step 3: Adopt a transfer learning mechanism for online adaptation to address the cold start problem of data at new construction sites; Step 4: Based on the predicted total energy, obtain the single pile driving energy through the pile driving strategy and drive the pile; Step 5: Based on the SHAP algorithm, perform interpretability analysis on the prediction model and output the feature importance ranking and decision basis.

2. The intelligent pile driver hammer impact energy optimization control method based on machine learning according to claim 1, characterized in that, The collected data includes: Operating parameters: current pile penetration depth, hammer type parameters, pile diameter, and wall thickness; Energy sequence: current and previous history Energy of each hammer blow Response sequence: The penetration sequence corresponding to the historical hammer blows; Geological characteristics: soil layer type and hardness.

3. The intelligent pile driver hammer impact energy optimization control method based on machine learning according to claim 1, characterized in that, The prediction model uses a long short-term memory network, a Transformer, or an ensemble tree model.

4. The intelligent pile driver hammer impact energy optimization control method based on machine learning according to claim 1, characterized in that, The pile driving strategy in step 3 is to calculate the predicted total energy value by using different single hammer blow energies under the current working conditions within the single safe hammer blow energy. The single hammer blow energy corresponding to the minimum total energy value is taken as the pile driving strategy, and pile driving is carried out using this single pile driving energy.

5. The intelligent pile driver hammer impact energy optimization control method based on machine learning according to claim 1, characterized in that, The prediction model uses historical construction site data during the training phase and employs mean squared error or mean absolute error as the loss function for supervised learning.

6. The intelligent pile driver hammer impact energy optimization control method based on machine learning according to claim 1, characterized in that, The prediction model employs a transfer learning mechanism for online adaptation to address the cold start problem of new construction site data.

7. The intelligent pile driver hammer impact energy optimization control method based on machine learning according to claim 6, characterized in that, Employing transfer learning mechanisms: Pre-training phase: Train the basic model using pile driving data from multiple historical construction sites; Fine-tuning phase: Before construction begins at the new site, test pile data is collected as target domain data to fine-tune the foundation model.

8. A system for implementing the intelligent pile driver hammer impact energy optimization control method based on machine learning as described in claim 1, characterized in that, include: Sensing unit, control unit; The sensing unit is used to collect vibration signals during the hammering process, measure the pile penetration depth in real time, monitor the status of the hydraulic system and calculate the actual hammering energy, and collect environmental and working conditions. The control unit, using a PLC or industrial computer, is responsible for data acquisition, feature extraction, model prediction and optimization calculation, and receives energy commands from the controller to adjust the hammering energy of the hydraulic hammer in real time.

9. A computer device comprising a memory and a processor, the memory being electrically connected to the processor, the memory storing a computer program, characterized in that: When the computer program is executed by the processor, it causes the processor to implement the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the processor implements the method as described in any one of claims 1 to 8.

Images