Phase change regulation and control method and system for engine protection micro-aerobic environment

Abstract

The invention discloses an engine protection micro-aerobic environment phase change regulation and control method and system, and relates to the technical field of laser cladding machining, the phase change regulation and control method comprises the steps that S1, a target process window is set, and the target process window comprises a target cooling rate or a target cooling rate interval; the target cooling rate is pre-determined according to the expected metallographic structure and mechanical property of the engine protective layer. A target process window is set, multiband spectral radiation intensity data of a solidification area at the tail of a molten pool is collected in real time in combination with a multispectral sensing unit coaxially integrated with a laser head, and the multiband spectral radiation intensity data is preprocessed and then input into a pre-trained cooling rate calculation model to obtain a real-time cooling rate. And the controller calculates the deviation and dynamically adjusts the process parameters to form closed-loop control, so that the problem that the traditional open-loop control cannot respond to the cooling rate change of the molten pool in real time is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of laser cladding technology, specifically to a method and system for controlling phase change in the micro-oxygen environment of an engine enclosure. Background Technology

[0002] Laser cladding technology, as an advanced surface modification and repair method, has been widely used in high-end industrial fields such as aerospace, automotive manufacturing, and energy equipment due to its significant advantages, including high coating bonding strength, low dilution rate, and high material utilization. This technology uses a high-energy laser beam to melt pre-placed or synchronously fed alloy powder, forming a molten pool that rapidly solidifies, thereby creating a cladding layer with special properties on the substrate surface. This plays an irreplaceable role in improving the service life and reliability of key components.

[0003] In laser cladding, the phase transformation behavior of the molten pool is the core factor determining the final performance of the cladding layer. The phase transformation result is directly related to the metallographic structure of the cladding layer, such as grain size, phase composition, and precipitate distribution. These microstructural features further determine the mechanical and performance properties of the cladding layer. Therefore, achieving precise control over the phase transformation process is key to obtaining a stable and high-performance cladding layer.

[0004] Currently, the industry mostly adopts an open-loop control strategy for phase transition control during laser cladding. This involves pre-setting process parameters such as laser power, scanning speed, and powder supply based on past experience or theoretical calculations, and keeping these parameters constant throughout the entire processing. However, the actual processing environment is extremely complex and affected by various uncertainties: for example, the heat dissipation conditions vary in different areas of the workpiece, which may lead to localized heat accumulation or abnormal heat dissipation rates; the powder supply system may experience instantaneous supply deviations due to mechanical vibrations, air pressure fluctuations, etc.; and even the energy stability of the laser beam itself may be affected by fluctuations in the external power grid.

[0005] The combined effect of these factors can cause key parameters such as the actual temperature field distribution and cooling rate of the molten pool to deviate from the ideal design values. Excessive cooling can lead to the formation of supersaturated solid solutions, amorphous structures, or microcracks in the cladding layer; while excessively slow cooling can promote coarse grains and the precipitation of brittle phases, reducing the overall performance of the cladding layer. Ultimately, this results in significant uncertainty and fluctuation in the solidified metallographic structure, making it difficult to consistently achieve the desired performance and severely restricting the further application of laser cladding technology in fields requiring high precision and high reliability.

[0006] Especially in the laser cladding repair of critical components such as engine guards, the presence of micro-oxygen conditions in the working environment places more stringent demands on the control of the cooling rate of the molten pool. Traditional open-loop control methods cannot sense and respond to changes in the cooling rate during the solidification process of the molten pool in real time, making it difficult to accurately control the phase transformation process and ensuring the stability of the metallographic structure and mechanical properties of the cladding layer. Summary of the Invention

[0007] To address the shortcomings of existing technologies, this invention provides a method and system for controlling phase change in the micro-oxygen environment of an engine enclosure. This method and system have the advantages of achieving precise phase change control through real-time closed-loop control of the cooling rate, thereby improving the stability of the metallographic structure and mechanical properties of the cladding layer.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a method for controlling the phase change of the micro-oxygen environment in an engine enclosure, wherein the phase change control method includes:

[0009] Step S1: Set the target process window, which includes a target cooling rate or a target cooling rate range; the target cooling rate is predetermined based on the desired metallographic structure and mechanical properties of the engine guard layer.

[0010] Step S2: During the laser cladding process, the multi-band spectral radiation intensity data of the solidification area at the tail of the molten pool at a preset time resolution is collected in real time by the multi-spectral sensing unit coaxially integrated with the laser head, forming a real-time spectral data stream.

[0011] The multi-band spectral radiance data includes narrowband spectral information at least three different center wavelengths;

[0012] Step S3: Input the real-time spectral data stream into a pre-trained cooling rate calculation model, perform online inference calculation, and obtain the real-time cooling rate that is synchronized with the real-time spectral data stream.

[0013] Step S4: Set up a controller, which calculates the deviation signal between the real-time cooling rate and the target cooling rate in real time;

[0014] Step S5: The controller generates a process parameter adjustment command based on the deviation signal and outputs it to the laser cladding equipment to dynamically adjust at least one process parameter in a closed-loop manner, so that the real-time cooling rate is maintained within the target process window, thereby realizing online control of the phase transition.

[0015] The process parameters include laser power or laser scanning speed.

[0016] Furthermore, in step S2, the multispectral sensing unit collects spectral radiation intensity data from multiple spatial sampling points within the solidification region to obtain spatial distribution information reflecting the temperature gradient at the solidification front.

[0017] Furthermore, the cooling rate calculation model is a machine learning model based on a neural network, and the step of inputting the real-time spectral data stream into the cooling rate calculation model includes:

[0018] Step A1: Preprocess each frame of multi-band spectral radiance data in the real-time spectral data stream, including noise filtering and normalization operations;

[0019] Step A2: Construct a spectral feature vector from the preprocessed multi-band spectral radiance data;

[0020] Step A3: The spectral feature vector is used as input and transmitted to the input layer of the neural network-based machine learning model;

[0021] Step A4: Calculate the predicted value of the real-time cooling rate at the output layer through forward propagation of the model.

[0022] Furthermore, the training process of the neural network-based machine learning model includes:

[0023] Constructing a training sample set: The training sample set includes N sets of training samples. Each set of training samples consists of a set of spectral feature vectors collected under specific process parameters and corresponding labels of actual cooling rates measured by high-precision thermocouples or infrared thermal imagers.

[0024] Construct a neural network model: set up an input layer, at least one hidden layer, and an output layer in parallel; the number of nodes in the input layer corresponds to the dimension of the spectral feature vector, and the output of the output layer is a single cooling rate value;

[0025] Define the loss function: The mean squared error function is used as the loss function of the model, and the optimization objective is to minimize the mean squared error between the model's predicted value and the actual cooling rate label;

[0026] Model training: The training sample set is imported into the constructed neural network model in batches for iterative training. The network weights are updated using the backpropagation algorithm and gradient descent optimizer until the value of the loss function converges or the preset maximum number of iterations is reached, thus obtaining the trained cooling rate calculation model.

[0027] Furthermore, the process of constructing the training sample set also includes a data augmentation step, which includes adding Gaussian white noise or performing random time shifts to the original spectral feature vector to expand the size and diversity of the training sample set.

[0028] Further, in step S4, the formula for calculating the deviation signal E(t) is:

[0029] E(t) = R_target - R_real(t),

[0030] Where R_target is the target cooling rate, and R_real(t) is the real-time cooling rate at time t.

[0031] Furthermore, in step S5, the controller uses a proportional-integral-derivative (PID) control algorithm to generate process parameter adjustment instructions based on the deviation signal E(t);

[0032] The method for calculating the adjustment amount ΔP(t) of the laser power is as follows:

[0033] ΔP(t)=Kp·E(t)+Ki·∫0tE(τ)dτ+Kd·dE(t) / dt,

[0034] Where Kp, Ki, and Kd are the proportional, integral, and differential coefficients, respectively.

[0035] Furthermore, the method also includes performing a calibration procedure before laser cladding begins to confirm the alignment accuracy of the multispectral sensing unit with the solidification area of ​​the molten pool and the stability of signal acquisition.

[0036] The present invention also provides an engine enclosure micro-oxygen environment phase change control system for implementing the phase change control method, the phase change control system comprising:

[0037] Process parameter setting module: used to set a target process window including the target cooling rate, and send the target cooling rate to the process control module;

[0038] Data acquisition module: includes a multispectral sensing unit coaxially integrated with the laser head, used to acquire multi-band spectral radiation intensity data of the solidification area at the tail of the molten pool in real time during laser cladding process;

[0039] Data processing and calculation module: Coupled to the data acquisition module, it has a built-in pre-trained cooling rate calculation model, which is used to receive and process the multi-band spectral radiation intensity data and calculate it into a real-time cooling rate.

[0040] Process control module: coupled to the data processing and calculation module and the process parameter setting module, used to receive the real-time cooling rate and the target cooling rate, calculate the deviation signal between the two, and generate process parameter adjustment instructions based on the deviation signal;

[0041] The execution adjustment module is connected to the process control module and controls the laser cladding equipment. It is used to adjust the laser power or laser scanning speed in real time according to the process parameter adjustment instructions.

[0042] Furthermore, the data processing and calculation module also includes a data preprocessor, which performs noise filtering and normalization operations on the original multi-band spectral radiance data, constructs the processed data into a spectral feature vector, and then transmits it to the cooling rate calculation model.

[0043] Beneficial effects

[0044] This invention sets a target process window and uses a multispectral sensing unit coaxially integrated with the laser head to collect multi-band spectral radiation intensity data of the solidification region at the tail of the molten pool in real time. After preprocessing, the data is input into a pre-trained cooling rate calculation model to obtain the real-time cooling rate. The controller then calculates the deviation and dynamically adjusts the process parameters to form a closed-loop control, which effectively solves the problem that traditional open-loop control cannot respond to changes in the cooling rate of the molten pool in real time. Specifically, the multispectral sensing unit collects spectral data from multiple spatial sampling points, enabling the acquisition of spatial distribution information reflecting the temperature gradient at the solidification front, thus improving the comprehensiveness and accuracy of temperature field monitoring. The neural network-based cooling rate calculation model, after data augmentation training, can accurately handle nonlinear relationships. Combined with data preprocessing steps, this further improves the accuracy and robustness of cooling rate prediction. The PID control algorithm dynamically adjusts the laser power or scanning speed based on the deviation signal, enabling rapid response to cooling rate fluctuations and maintaining it within the target process window in real time. This avoids problems such as oversaturated solid solutions, amorphous structures, or microcracks caused by excessively fast cooling rates, and grain coarsening and brittle phase precipitation caused by excessively slow cooling rates. This improves the uniformity and stability of the cladding layer's metallographic structure, thereby ensuring the mechanical and performance properties of the cladding layer. It is particularly suitable for laser cladding repair of critical components such as engine mounts, meeting the application requirements of high-precision and high-reliability fields. Attached Figure Description

[0045] Figure 1 This is a flowchart of the method of the present invention;

[0046] Figure 2 This is a system structure diagram of the present invention.

[0047] In the picture:

[0048] 101. Process parameter setting module; 201. Data acquisition module; 301. Process control module; 401. Process control module; 501. Execution adjustment module. Detailed Implementation

[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0050] Please see Figure 1 This invention proposes a method for controlling the phase change in the micro-oxygen environment of an engine enclosure. The phase change control method includes:

[0051] Step S1: Set the target process window, which includes the target cooling rate or the target cooling rate range; the target cooling rate is predetermined based on the desired metallographic structure and mechanical properties of the engine guard layer.

[0052] Step S2: During the laser cladding process, the multi-band spectral radiation intensity data of the solidification area at the tail of the molten pool at a preset time resolution is collected in real time by a multi-spectral sensing unit coaxially integrated with the laser head, forming a real-time spectral data stream; the multi-band spectral radiation intensity data includes narrowband spectral information of at least three different center wavelengths.

[0053] Step S3: Input the real-time spectral data stream into a pre-trained cooling rate calculation model, perform online inference calculation, and obtain the real-time cooling rate synchronized with the real-time spectral data stream;

[0054] S4: Set up a controller that calculates the deviation signal between the real-time cooling rate and the target cooling rate in real time.

[0055] Step S5: The controller generates process parameter adjustment instructions based on the deviation signal and outputs them to the laser cladding equipment. At least one process parameter is dynamically adjusted in a closed-loop manner to maintain the real-time cooling rate within the target process window, thereby achieving online control of the phase change.

[0056] Process parameters include laser power or laser scanning speed.

[0057] The target process window refers to a pre-set cooling rate or cooling rate range, which can be determined by combining experimental data with theoretical models. By analyzing the influence of different cooling rates on metallographic structure and mechanical properties, a mapping relationship between process parameters and performance can be established.

[0058] A multispectral sensing unit is an optical detection device coaxially integrated with a laser head. Specifically, it can be implemented by combining multiple narrowband filters with a photodetector. By collecting multi-band spectral radiation intensity data of the solidification region of the molten pool, it reflects the dynamic changes in the temperature field.

[0059] The cooling rate calculation model refers to a calculation model based on the relationship between spectral data and cooling rate. Specifically, it can be trained using a neural network model, which can output a predicted value of the real-time cooling rate by inputting multi-band spectral feature vectors.

[0060] A controller is a closed-loop control unit used to dynamically adjust process parameters. Specifically, it can be implemented using a proportional-integral-derivative algorithm. It generates adjustment commands by calculating the cooling rate deviation in real time, ensuring the response speed and stability of process parameter adjustments.

[0061] Dynamic adjustment of process parameters refers to closed-loop control of laser power or scanning speed. Specifically, it can be performed by using a servo motor or laser modulator. By changing the energy input or processing speed, the cooling rate of the molten pool is directly affected.

[0062] The core innovation of this invention lies in capturing the spectral characteristics of the solidification region of the molten pool in real time through multispectral sensing, combining the actual cooling rate with the cooling rate calculation model to infer the actual cooling rate online, and dynamically adjusting the process parameters based on closed-loop control to stabilize the cooling rate within the target process window, thereby achieving precise control of the phase transformation process.

[0063] The working process and principle of this scheme are as follows: The phase change control method for the micro-oxygen environment of the engine enclosure first sets a target process window, including a target cooling rate or a target cooling rate range. The target cooling rate is predetermined based on the desired metallographic structure and mechanical properties of the engine enclosure layer. During the laser cladding process, multi-band spectral radiation intensity data of the solidification region at the tail of the molten pool are collected in real time by a multi-spectral sensing unit coaxially integrated with the laser head, forming a real-time spectral data stream. The multi-band spectral radiation intensity data includes narrowband spectral information of at least three different center wavelengths to obtain more comprehensive temperature field distribution information.

[0064] The real-time spectral data stream is input into a pre-trained cooling rate calculation model for online inference calculation to obtain the real-time cooling rate synchronized with the real-time spectral data stream. The controller calculates the deviation signal between the real-time cooling rate and the target cooling rate in real time and generates process parameter adjustment commands based on the deviation signal. The adjustment commands are output to the laser cladding equipment to dynamically adjust process parameters such as laser power or laser scanning speed in a closed-loop manner, so that the real-time cooling rate is maintained within the target process window, thereby achieving online control of the phase transition.

[0065] This method, through real-time spectral acquisition and cooling rate calculation, combined with a closed-loop control strategy, can quickly respond to dynamic changes in the cooling state of the molten pool, adjust process parameters in a timely manner, effectively suppress the influence of environmental interference and process fluctuations on the phase transformation process, and ensure the uniformity and stability of the microstructure of the cladding layer.

[0066] As a preferred embodiment, this solution

[0067] In the laser cladding repair process of engine turbine blades, the target cooling rate was first set to 500°C / s based on the characteristics of the nickel-based superalloy matrix material and the desired oriented columnar crystal structure. The multispectral sensing unit employed three narrowband filters with center wavelengths of 800nm, 1000nm, and 1200nm, along with a photodetector, and a sampling frequency of 1kHz. The cooling rate calculation model used a three-layer fully connected neural network structure, with the input layer corresponding to the spectral intensity of the three bands, the hidden layer containing 20 neurons, and the output layer representing a single cooling rate value.

[0068] The controller employs a PID algorithm to calculate process parameter adjustments based on the deviation between the real-time cooling rate and the target cooling rate. When a sudden increase in the cooling rate in the blade tenon region is detected, the controller rapidly reduces the laser power or slows down the scanning speed to maintain the cooling rate within the target range. Throughout the process, the system continuously monitors and adjusts to ensure that a uniformly oriented columnar crystal structure is formed in each region of the cladding layer.

[0069] Furthermore, the multispectral sensing unit collects spectral radiation intensity data from multiple spatial sampling points within the solidification region to obtain spatial distribution information reflecting the temperature gradient at the solidification front.

[0070] The multispectral sensing unit employs an array of sensors or a scanning mechanism to spatially discretize the solidification region, with each sampling point corresponding to a specific location at the solidification front. Spectral radiation intensity data covers multiple measurement points distributed laterally and longitudinally within the solidification region, with the spacing between adjacent sampling points dynamically adjusted according to the molten pool size. By simultaneously acquiring spectral radiation intensity data from different spatial locations, a two-dimensional temperature gradient distribution matrix is ​​formed.

[0071] Specifically, during laser cladding, the heat transfer direction in the solidification region at the tail of the molten pool is closely related to its spatial location. A multispectral sensing unit divides the solidification region into a grid using a preset sampling density, for example, setting five equally spaced sampling points along the width of the molten pool and three sampling layers vertically. The spectral radiation intensity data of each sampling point is synchronously captured by a multi-channel photodetector, and its spatial coordinate information is recorded. As the solidification front of the molten pool moves, the multispectral sensing unit drives an optical probe via a servo motor to follow and scan, ensuring that the sampling point always covers the current solidification region. By analyzing the differences in spectral radiation intensity at different spatial locations, the temperature gradient distribution curve of the solidification front can be reconstructed. This spatial distribution information, after being input into the cooling rate calculation model, can correct calculation errors caused by sudden changes in local heat dissipation conditions. For example, when a sampling point experiences abnormally high temperatures due to substrate impurities, the model can automatically interpolate and compensate based on data from adjacent sampling points, thereby improving the accuracy of real-time cooling rate calculations.

[0072] As a preferred embodiment, this solution

[0073] A multispectral sensing unit acquires spectral radiance data from multiple spatial sampling points within the solidification region to obtain spatial distribution information reflecting the temperature gradient at the solidification front. Specifically, the multispectral sensing unit includes a high-speed CCD camera and a set of narrowband optical filters. The CCD camera's field of view covers the entire solidification region at the tail of the molten pool, with a resolution of 1920×1080 pixels. The optical filter set contains three narrowband filters with center wavelengths of 800nm, 1000nm, and 1200nm, each with a bandwidth of 10nm. The CCD camera uses a time-division multiplexing method, sequentially switching the three filters in each sampling cycle to acquire spectral images of the corresponding wavelengths. The sampling frequency is set to 100Hz, meaning a complete three-band spectral acquisition is completed every 10ms. During image processing, the solidification region is divided into a 10×10 grid, and the average pixel intensity within each grid is used as the spectral radiance of that spatial sampling point. Thus, each sampling yields 300 data points (10×10×3), forming a multidimensional feature vector reflecting the spatial distribution of the temperature gradient at the solidification front.

[0074] Furthermore, the cooling rate calculation model is a neural network-based machine learning model. Real-time spectral data streams are input into the cooling rate calculation model, including:

[0075] Step A1: Preprocess each frame of multi-band spectral radiance data in the real-time spectral data stream, including noise filtering and normalization operations;

[0076] Step A2: Construct a spectral feature vector from the preprocessed multi-band spectral radiance data;

[0077] Step A3: The spectral feature vector is used as input and transmitted to the input layer of the neural network-based machine learning model;

[0078] Step A4: Calculate the predicted value of the real-time cooling rate at the output layer through forward propagation of the model.

[0079] The noise filtering employs a sliding window averaging method with a window width of 5 data points, effectively suppressing high-frequency noise. Normalization linearly maps the intensity values ​​of each band to the 0-1 interval, eliminating dimensional differences. The spectral feature vector is constructed by arranging the intensities of multiple bands at the same time point in ascending order of wavelength, forming a vector with a dimension equal to the number of bands. The neural network model uses a fully connected structure, with the number of input layer nodes equal to the number of bands, two hidden layers (each containing 32 nodes), ReLU activation, and a single-node linear output.

[0080] Specifically, each frame of spectral data first undergoes a moving average filter to remove random noise caused by electromagnetic interference or detector jitter, improving signal smoothness by approximately 40%. Normalization is then performed by calculating the ratio of each band's intensity to its historical maximum value, eliminating intensity baseline drift caused by laser power fluctuations. The constructed spectral feature vector integrates multi-dimensional spectral information into a standardized data structure. After being input into the neural network, the mapping relationship between spectral intensity and cooling rate is extracted through nonlinear transformations in the hidden layers. During forward propagation, the input layer receives vector data, the first hidden layer calculates the weighted sum of each node and activates it, the second hidden layer further extracts higher-order features, and the final output layer maps the features to cooling rate values. This process enables the model to complete a single inference within 0.5 milliseconds, with prediction errors controlled within ±3%.

[0081] Furthermore, the training process for the cooling rate calculation model includes:

[0082] Constructing a training sample set: The training sample set includes N sets of training samples. Each set of training samples consists of a set of spectral feature vectors collected under specific process parameters and corresponding labels of actual cooling rates measured by high-precision thermocouples or infrared thermal imagers.

[0083] Construct a neural network model: set up an input layer, at least one hidden layer, and an output layer in parallel; the number of nodes in the input layer corresponds to the dimension of the spectral feature vector, and the output of the output layer is a single cooling rate value;

[0084] Define the loss function: The mean squared error function is used as the loss function of the model, and the optimization objective is to minimize the mean squared error between the model's predicted value and the actual cooling rate label;

[0085] Model training: The training sample set is imported into the constructed neural network model in batches for iterative training. The network weights are updated using the backpropagation algorithm and gradient descent optimizer until the value of the loss function converges or the preset maximum number of iterations is reached, thus obtaining the trained cooling rate calculation model.

[0086] In constructing the training sample set, the spectral feature vectors of each training sample are acquired by a multispectral sensing unit under a specific combination of laser power and scanning speed. The actual cooling rate label is simultaneously measured by a contact thermocouple or a non-contact infrared thermal imager. In the data augmentation step, the amplitude of Gaussian white noise is controlled within 5% of the original spectral signal intensity, and the range of random time offset does not exceed ±50 milliseconds. The hidden layer of the neural network model adopts a fully connected structure, the activation function is the ReLU function, and the number of hidden layer nodes is set to 1.5 times the number of input layer nodes. When calculating the loss function, the sample size of each batch is set to 32, the initial learning rate of the gradient descent optimizer is 0.001, and an exponential decay strategy is adopted.

[0087] Specifically, in the training sample construction phase, a 10-second continuous spectral data stream was acquired using a fixed combination of process parameters: laser power of 1500W and scanning speed of 8mm / s. Simultaneously, a 100Hz infrared thermal imager was used to measure the temperature change curve at the tail of the molten pool. The actual cooling rate label was obtained by calculating the first derivative of the temperature-time curve, forming a set of training samples. In the data preprocessing phase, the raw spectral data underwent moving average filtering with a window width of 5 data points, followed by max-min normalization. During model training, 32 samples were input per batch. During forward propagation, the ReLU function of the hidden layer was set to zero to prevent gradient vanishing, and a linear activation function was used in the output layer. During backpropagation, the weight update was calculated based on the partial derivative of the loss function with respect to the network parameters, and the learning rate decreased to 0.9 times its original value every 1000 iterations. When the number of training iterations reached 5000, the mean squared error on the validation set decreased to below 0.15℃ / s, indicating that the model has a reliable cooling rate prediction capability.

[0088] Furthermore, the process of constructing the training sample set also includes a data augmentation step, which involves adding Gaussian white noise or performing random time shifts to the original spectral feature vectors to expand the size and diversity of the training sample set.

[0089] The data augmentation step introduces Gaussian white noise to simulate electromagnetic interference or sensor noise that may exist during actual data acquisition, improving the model's adaptability to noisy environments. Random time offset, through small displacements of the spectral feature vector over time, simulates the temporal differences in the dynamic changes of temperature gradients during molten pool solidification, enhancing the model's robustness to temporal signal fluctuations. These two augmentation methods expand the coverage of training samples from two dimensions: spatial noise immunity and temporal feature generalization, respectively, creating a complementary effect.

[0090] Specifically, when constructing the training sample set, the original spectral feature vector is injected with Gaussian white noise to generate derived samples containing different signal-to-noise ratio levels. This enables the model to learn the ability to accurately extract spectral features under noise interference. Simultaneously, a random time offset operation is applied to the original spectral feature vector to generate samples with time-series phase differences, simulating the non-uniformity of temperature gradient changes in the solidification region of the molten pool during actual processing. Through these two data augmentation methods, the size of the training sample set is expanded exponentially, and the distribution of sample features more closely reflects the complexity and diversity of real-world operating conditions. The cooling rate calculation model trained in this way can effectively suppress noise interference, adapt to different temporal dynamic characteristics, significantly improve the generalization performance of cooling rate prediction, and ensure the stability of the closed-loop control system during online inference.

[0091] As a preferred embodiment, this solution

[0092] During the construction of the training sample set, data augmentation steps are introduced to expand its size and diversity. Specifically, Gaussian white noise or random time shifts are added to the original spectral feature vector. For example, Gaussian white noise with a mean of 0 and a standard deviation of 0.01 can be superimposed on each component of the original spectral feature vector; or the original spectral feature vector can be randomly shifted forward or backward by 1-5 sampling points on the time axis. In this way, signal noise and time delays that may occur in actual working conditions can be simulated, improving the model's robustness to these interference factors.

[0093] Data augmentation steps can be performed according to the following process:

[0094] Generate 5-10 enhanced samples for each original spectral feature vector.

[0095] For Gaussian white noise enhancement, a noise vector with the same dimension as the original vector is generated using a random number generator and then added to the original vector.

[0096] For time offset enhancement, a random offset is selected, and the original vector is shifted by that offset number of sampling points.

[0097] The enhanced samples are merged with the original samples to form an expanded training sample set.

[0098] Furthermore, the formula for calculating the deviation signal is: E(t) = R_target - R_real(t), where R_target is the target cooling rate and R_real(t) is the real-time cooling rate at time t.

[0099] R_target is set based on the pre-determined metallographic structure and mechanical properties of the engine guard layer, while R_real(t) is obtained through online inference using a cooling rate calculation model. The formula uses the direct difference between the real-time cooling rate and the target value to avoid introducing complex calculation steps and ensure the real-time performance of the control system. The sign of the difference signal directly reflects the direction of deviation of the cooling rate from the target, providing a clear input basis for subsequent control algorithms.

[0100] Specifically, during laser cladding, a multispectral sensing unit collects spectral data from the solidification region of the molten pool. After preprocessing, this data is input into a cooling rate calculation model, outputting a real-time cooling rate R_real(t). This rate is instantaneously compared with a preset R_real(t) using a subtractor, generating a deviation signal E(t). This signal is transmitted to the controller as an input parameter for a proportional-integral-derivative (PID) control algorithm, dynamically adjusting the laser power or scanning speed. Through the linearity of the subtraction operation, the system can quickly respond to instantaneous fluctuations in the cooling rate, eliminating the lag caused by traditional empirical threshold judgments and improving the accuracy and stability of closed-loop control.

[0101] Assuming the target cooling rate R_target is set to 500°C / s, at a certain time t, the real-time cooling rate R_real(t) calculated by the cooling rate calculation model from the real-time spectral data collected by the multispectral sensing unit is 480°C / s. Therefore, the deviation signal E(t) at that time can be calculated as 500 - 480 = 20°C / s.

[0102] Furthermore, the controller can generate corresponding process parameter adjustment commands based on the calculated deviation signal E(t). For example, when E(t) is positive, it indicates that the actual cooling rate is lower than the target value, and the controller may generate commands to increase the laser power or decrease the scanning speed to increase the molten pool temperature and accelerate the cooling rate. Conversely, when E(t) is negative, the controller may generate commands to decrease the laser power or increase the scanning speed to decrease the molten pool temperature and slow down the cooling rate.

[0103] Furthermore, a proportional-integral-derivative (PID) control algorithm is used to generate process parameter adjustment commands based on the deviation signal. The adjustment amount of the laser power is calculated by superimposing proportional, integral, and derivative terms.

[0104] The proportional term consists of the product of the real-time deviation and the proportional coefficient, used for rapid response to the current deviation; the integral term consists of the product of the accumulated historical deviation and the integral coefficient, used to eliminate steady-state error; and the derivative term consists of the product of the rate of change of deviation and the derivative coefficient, used to suppress overshoot. The proportional coefficient, integral coefficient, and derivative coefficient are determined through experimental calibration or system identification methods. For example, during calibration, the combination of coefficients is adjusted through step response testing to bring the system to a critical damped state.

[0105] Specifically, the controller collects the real-time cooling rate at every moment, calculates the deviation from the target value, and then performs a weighted summation of the current deviation value, the historical integral value of the deviation, and the rate of change of the deviation to generate the laser power adjustment. For example, when the real-time cooling rate is lower than the target value, the deviation value is positive, and the controller increases the laser power to raise the molten pool temperature, thereby slowing down the cooling rate. If the deviation persists, the integral term gradually accumulates, further enhancing the adjustment force. If the deviation changes rapidly, the derivative term predicts the trend in advance and suppresses power abrupt changes. Through the synergistic effect of these three factors, the dynamic error of the cooling rate is effectively suppressed, and the actual value stably converges to within the target process window. Compared with single proportional control, the response speed is improved by about 30%, and the steady-state error is reduced to within ±2%.

[0106] As a preferred embodiment, this solution

[0107] The controller employs a proportional-integral-derivative (PID) control algorithm to generate process parameter adjustment commands based on the deviation signal E(t). The calculation method for the laser power adjustment ΔP(t) is as follows:

[0108] ΔP(t)=Kp·E(t)+Ki·∫0tE(τ)dτ+Kd·dE(t) / dt,

[0109] Where Kp, Ki, and Kd are the proportional, integral, and differential coefficients, respectively.

[0110] In practice, the parameters of the PID controller can be set according to the actual process requirements. For example, in a laser cladding system, Kp=0.5, Ki=0.1, and Kd=0.05 can be set. When the real-time cooling rate deviates from the target value, the controller will calculate the required laser power adjustment based on the PID algorithm.

[0111] Suppose that at a certain time t, the detected real-time cooling rate is 20°C / s lower than the target value, i.e., E(t) = 20°C / s. The controller will immediately calculate the adjustment amount of the laser power:

[0112] ΔP(t)=0.5·20+0.1·∫0t20dτ+0.05·d(20) / dt,

[0113] The calculations showed that the laser power needed to be increased. The controller then sent a command to the laser to increase the power, causing the real-time cooling rate to gradually approach the target value. This process continues until the deviation signal approaches zero, achieving precise control of the cooling rate.

[0114] Furthermore, before laser cladding begins, a calibration procedure is performed to confirm the alignment accuracy between the multispectral sensing unit and the solidification area of ​​the molten pool, as well as the stability of signal acquisition.

[0115] The calibration procedure includes the following steps: First, with the laser cladding equipment off, a calibration template with preset markers is fixed at the expected position in the solidification area of ​​the molten pool using a mechanical positioning device. Second, the optical acquisition module of the multispectral sensing unit is activated, enabling it to scan the markers on the calibration template from multiple angles to obtain the spatial coordinate data of the markers. Next, by comparing the actual acquired coordinate data with the preset theoretical coordinate values, the deviation between the optical axis of the multispectral sensing unit and the coaxiality of the laser head is calculated, and the servo motor is driven to adjust the installation angle of the multispectral sensing unit until the deviation is less than a preset threshold. Finally, under simulated conditions where the laser power maintains low energy output, the spectral signal of the simulated molten pool area is continuously acquired, and the standard deviation of the signal intensity is statistically analyzed. If it exceeds the allowable range, the optical focusing parameters are automatically adjusted or an alarm is triggered.

[0116] Specifically, the calibration procedure ensures that the optical axis of the multispectral sensing unit is strictly coaxial with the laser cladding processing area by physically calibrating the spatial positioning of the template, avoiding spectral signal acquisition angle deviation caused by installation misalignment. Under low-power simulation conditions, the system continuously acquires spectral data and analyzes its stability, identifying potential interference sources such as optical lens contamination, sensor aging, or circuit noise. When signal fluctuations exceed the preset tolerance, the system automatically optimizes focusing parameters or prompts maintenance personnel for repair, thus eliminating abnormal factors in the signal acquisition process before formal processing. This procedure, through a dual mechanism of mechanical positioning calibration and dynamic signal stability verification, ensures that the spectral data output by the multispectral sensing unit during subsequent cladding processes has sufficient spatial alignment accuracy and temporal stability, providing reliable input for the cooling rate calculation model and thus ensuring the dynamic response accuracy of the closed-loop control system.

[0117] In a preferred embodiment, during the startup phase of the laser cladding equipment, a pre-fabricated calibration test plate is used as the processing substrate during the calibration procedure. A high-reflectivity marker array is pre-placed on the surface of the calibration test plate, and the spatial coordinates of the marker points are pre-calibrated using a laser tracker. Subsequently, the laser head is controlled to emit a visible-band auxiliary beam in low-power mode, while simultaneously activating the multispectral sensing unit to acquire reflected light signals. An image processing algorithm identifies the positional offset of the marker points in the multispectral imaging, and the spatial deviation between the laser head's optical axis and the optical center axis of the multispectral sensing unit is calculated. If the deviation exceeds a preset threshold, a mechanical adjustment mechanism is automatically triggered to compensate for and correct the installation angle of the multispectral sensing unit until the offset converges to an allowable range. Further, after correction, a molten pool simulation experiment is performed on the test plate surface with constant laser power, continuously acquiring spectral signals and calculating their signal-to-noise ratio (SNR). When the SNR is lower than a set standard, the integration time or gain parameter of the multispectral sensing unit is automatically adjusted until the signal fluctuation amplitude stabilizes within the allowable threshold range.

[0118] Please see Figure 2 The present invention also proposes an engine enclosure micro-oxygen environment phase change control system for implementing the above-mentioned phase change control method. The phase change control system includes:

[0119] Process parameter setting module 101: used to set the target process window including the target cooling rate, and send the target cooling rate to the process control module;

[0120] Data acquisition module 201: includes a multispectral sensing unit coaxially integrated with the laser head, used to acquire multi-band spectral radiation intensity data of the solidification area at the tail of the molten pool in real time during the laser cladding process;

[0121] Data processing and calculation module 301: Coupled to the data acquisition module, it has a built-in pre-trained cooling rate calculation model, which is used to receive and process multi-band spectral radiation intensity data and calculate it into real-time cooling rate.

[0122] Process control module 401: Coupled to the data processing and calculation module and the process parameter setting module, it is used to receive the real-time cooling rate and the target cooling rate, calculate the deviation signal between the two, and generate process parameter adjustment instructions based on the deviation signal.

[0123] The adjustment module 501 is connected to the process control module and controls the laser cladding equipment. It is used to adjust the laser power or laser scanning speed in real time according to the process parameter adjustment instructions.

[0124] Specifically, the process parameter setting module 101 receives the target cooling rate input by the user and transmits the value to the process control module.

[0125] The data acquisition module 201 uses a spectral sensor coaxially integrated into the laser head to capture multi-band spectral data of the solidification area at the tail of the molten pool at a sampling rate of 500 frames per second during the laser cladding process. Its spectral coverage includes the visible light band 650nm, the near-infrared band 850nm and the short-wave infrared band 1450nm.

[0126] The data preprocessor of the data processing and calculation module 301 uses wavelet transform algorithm to filter out high-frequency noise. After normalizing the original spectral data, it combines the light intensity values ​​of the three bands into a three-dimensional feature vector, which is then input into the trained neural network model to output the real-time cooling rate.

[0127] The process control module 401 uses an incremental PID algorithm to generate adjustment commands based on the deviation between the real-time cooling rate and the target value.

[0128] The adjustment module 501 dynamically adjusts the laser power within the range of 500-2000W by adjusting the current drive signal of the laser, so that the cooling rate is stabilized within the target range.

[0129] Furthermore, the data processing and calculation module also includes a data preprocessor, which performs noise filtering and normalization operations on the original multi-band spectral radiance data, and constructs the processed data into spectral feature vectors before transmitting them to the cooling rate calculation model.

[0130] The noise filtering operation employs moving average filtering or wavelet thresholding algorithms to suppress high-frequency interference signals by setting a cutoff frequency or noise energy threshold. The normalization operation linearly scales the spectral intensity data of each band according to a preset benchmark value, eliminating dimensional differences between different bands. When constructing the spectral feature vector, the normalized data of multiple bands at the same time point are arranged in wavelength order into a one-dimensional array, forming a vector structure matching the number of nodes in the model input layer. For example, when the multispectral sensing unit acquires narrowband spectra at three center wavelengths, the preprocessed data is constructed into a vector containing three elements, each corresponding to the normalized intensity value of a band.

[0131] Specifically, in the data preprocessor, the raw spectral data first passes through a Gaussian low-pass filter to remove high-frequency fluctuations caused by electromagnetic interference or sensor noise. Subsequently, the data for each band is divided by a preset maximum intensity reference value, compressing the amplitude range to between 0 and 1. After these operations, the normalized data for the three bands are arranged as a vector in ascending order of wavelength and input to the input layer of the cooling rate calculation model. This process ensures that the input data received by the model has uniform dimensions and a stable signal-to-noise ratio, avoiding calculation errors caused by fluctuations in the quality of the raw data. For example, when a band experiences a momentary amplitude jump due to environmental interference, the noise filtering step effectively suppresses outliers, while the normalization step eliminates sensitivity differences between different sensor channels, enabling the model to perform inference calculations based on consistent data characteristics and improving the accuracy of real-time cooling rate predictions.

[0132] In summary, this invention sets a target process window and uses a multispectral sensing unit coaxially integrated with the laser head to collect multi-band spectral radiation intensity data of the solidification region at the tail of the molten pool in real time. After preprocessing, the data is input into a pre-trained cooling rate calculation model to obtain the real-time cooling rate. The controller then calculates the deviation and dynamically adjusts the process parameters to form a closed-loop control, which effectively solves the problem that traditional open-loop control cannot respond to changes in the cooling rate of the molten pool in real time. Specifically, the multispectral sensing unit collects spectral data from multiple spatial sampling points, enabling the acquisition of spatial distribution information reflecting the temperature gradient at the solidification front, thus improving the comprehensiveness and accuracy of temperature field monitoring. The neural network-based cooling rate calculation model, after data augmentation training, can accurately handle nonlinear relationships. Combined with data preprocessing steps, this further improves the accuracy and robustness of cooling rate prediction. The PID control algorithm dynamically adjusts the laser power or scanning speed based on the deviation signal, enabling rapid response to cooling rate fluctuations and maintaining it within the target process window in real time. This avoids problems such as oversaturated solid solutions, amorphous structures, or microcracks caused by excessively fast cooling rates, and grain coarsening and brittle phase precipitation caused by excessively slow cooling rates. This improves the uniformity and stability of the cladding layer's metallographic structure, thereby ensuring the mechanical and performance properties of the cladding layer. It is particularly suitable for laser cladding repair of critical components such as engine mounts, meeting the application requirements of high-precision and high-reliability fields.

[0133] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0134] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for controlling phase change in the micro-oxygen environment of an engine enclosure, characterized in that, The phase transition control method includes: Step S1: Set the target process window, which includes a target cooling rate or a target cooling rate range; the target cooling rate is predetermined based on the desired metallographic structure and mechanical properties of the engine guard layer. Step S2: During the laser cladding process, the multi-band spectral radiation intensity data of the solidification area at the tail of the molten pool at a preset time resolution is collected in real time by the multi-spectral sensing unit coaxially integrated with the laser head, forming a real-time spectral data stream. The multi-band spectral radiance data includes narrowband spectral information at least three different center wavelengths; Step S3: Input the real-time spectral data stream into a pre-trained cooling rate calculation model, perform online inference calculation, and obtain the real-time cooling rate that is synchronized with the real-time spectral data stream. Step S4: Set up a controller, which calculates the deviation signal between the real-time cooling rate and the target cooling rate in real time; Step S5: The controller generates a process parameter adjustment command based on the deviation signal and outputs it to the laser cladding equipment to dynamically adjust at least one process parameter in a closed-loop manner, so that the real-time cooling rate is maintained within the target process window, thereby realizing online control of the phase transition. The process parameters include laser power or laser scanning speed.

2. The phase transition control method according to claim 1, characterized in that, In step S2, the multispectral sensing unit collects spectral radiation intensity data from multiple spatial sampling points within the solidification region to obtain spatial distribution information reflecting the temperature gradient at the solidification front.

3. The phase transition control method according to claim 1, characterized in that, The cooling rate calculation model is a machine learning model based on a neural network. The step of inputting the real-time spectral data stream into the cooling rate calculation model includes: Step A1: Preprocess each frame of multi-band spectral radiance data in the real-time spectral data stream, including noise filtering and normalization operations; Step A2: Construct a spectral feature vector from the preprocessed multi-band spectral radiance data; Step A3: The spectral feature vector is used as input and transmitted to the input layer of the neural network-based machine learning model; Step A4: Calculate the predicted value of the real-time cooling rate at the output layer through forward propagation of the model.

4. The phase transition control method according to claim 3, characterized in that, The training process of the neural network-based machine learning model includes: Constructing a training sample set: The training sample set includes N sets of training samples. Each set of training samples consists of a set of spectral feature vectors collected under specific process parameters and corresponding labels of actual cooling rates measured by high-precision thermocouples or infrared thermal imagers. Construct a neural network model: set up an input layer, at least one hidden layer, and an output layer in parallel; the number of nodes in the input layer corresponds to the dimension of the spectral feature vector, and the output of the output layer is a single cooling rate value; Define the loss function: The mean squared error function is used as the loss function of the model, and the optimization objective is to minimize the mean squared error between the model's predicted value and the actual cooling rate label; Model training: The training sample set is imported into the constructed neural network model in batches for iterative training. The network weights are updated using the backpropagation algorithm and gradient descent optimizer until the value of the loss function converges or the preset maximum number of iterations is reached, thus obtaining the trained cooling rate calculation model.

5. The phase transition control method according to claim 4, characterized in that, The process of constructing the training sample set also includes a data augmentation step, which includes adding Gaussian white noise or performing random time shifts to the original spectral feature vectors to expand the size and diversity of the training sample set.

6. The phase transition control method according to claim 1, characterized in that, In step S4, the formula for calculating the deviation signal E(t) is: E(t) = R_target - R_real(t), Where R_target is the target cooling rate, and R_real(t) is the real-time cooling rate at time t.

7. The phase transition control method according to claim 6, characterized in that, In step S5, the controller uses a proportional-integral-derivative (PID) control algorithm to generate process parameter adjustment instructions based on the deviation signal E(t); The method for calculating the adjustment amount ΔP(t) of the laser power is as follows: ΔP(t)=Kp·E(t)+Ki·∫0tE(τ)dτ+Kd·dE(t) / dt, Where Kp, Ki, and Kd are the proportional, integral, and differential coefficients, respectively.

8. The phase transition control method according to claim 1, characterized in that, The method also includes executing a calibration procedure before laser cladding begins to confirm the alignment accuracy of the multispectral sensing unit with the solidification area of ​​the molten pool and the stability of signal acquisition.

9. A phase change control system for an engine enclosure micro-oxygen environment, used to implement the phase change control method as described in any one of claims 1-8, characterized in that, The phase transition control system includes: Process parameter setting module (101): used to set a target process window including the target cooling rate, and send the target cooling rate to the process control module; Data acquisition module (201): includes a multispectral sensing unit coaxially integrated with the laser head, used to acquire multi-band spectral radiation intensity data of the solidification area at the tail of the molten pool in real time during the laser cladding process; Data processing and calculation module (301): Coupled to the data acquisition module, it has a built-in pre-trained cooling rate calculation model, which is used to receive and process the multi-band spectral radiation intensity data and calculate it into a real-time cooling rate. Process control module (401): Coupled to the data processing and calculation module and the process parameter setting module, used to receive the real-time cooling rate and the target cooling rate, calculate the deviation signal between the two, and generate process parameter adjustment instructions based on the deviation signal; The execution adjustment module (501) is connected to the process control module and controls the laser cladding equipment. It is used to adjust the laser power or laser scanning speed in real time according to the process parameter adjustment instructions.

10. The phase change control system according to claim 9, characterized in that, The data processing and calculation module further includes a data preprocessor, which performs noise filtering and normalization operations on the original multi-band spectral radiance data, constructs the processed data into spectral feature vectors, and then transmits them to the cooling rate calculation model.

Images