Laser additive manufacturing molten pool temperature feedforward control method

Abstract

The invention belongs to the technical field of laser additive manufacturing process control, and particularly relates to a temperature feedforward control method for a laser additive manufacturing molten pool. Comprising the following steps: step 1, acquiring a surface temperature field of an Nth layer after solidification through a thermal imaging system as an initial state T0; step 2, acquiring a scanning path of an (N + 1) th layer from a CAM system, calculating an equivalent heat dissipation index of each point on the path, and generating an EHDI sequence; 3, inputting the T0 and EHDI sequences into a pre-trained LSTM prediction model, and predicting the dynamic change of the molten pool temperature; and step 4, based on a model prediction path integral control algorithm, optimizing and generating an optimal laser power feed-forward control sequence of N + 1 layers by taking minimization of a cost function including a temperature tracking error, a power change and a power size as a target. According to the method, the path information is quantified into the EHDI sequence, the thermal dynamic state is accurately predicted through the LSTM model, the MPPI controller can'predict 'thermal disturbance before scanning is started, the optimal power scheme is generated, and feedforward accurate control over the temperature of the molten pool is achieved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of process control of laser additive manufacturing (LAM), and particularly relates to a laser additive manufacturing molten pool temperature feedforward control method. BACKGROUND

[0002] Laser additive manufacturing technology, especially laser directed energy deposition, fabricates three-dimensional parts by melting metal powder layer by layer with high-energy laser beams. The core challenge of this process lies in its strong path dependence of thermal behavior: laser scanning paths cause heat to accumulate and dissipate non-uniformly in different areas of the part, causing the molten pool temperature to fluctuate dramatically, which directly affects the microstructure, density, residual stress and mechanical properties of the formed part, and is prone to defects such as pores and cracks.

[0003] Currently, traditional feedback control methods (such as PID control) are difficult to achieve effective control due to sensor delays and extremely fast process dynamic responses. Existing feedforward control methods rely on simplified analytical models or computationally expensive finite element simulations, and there is a problem of being unable to balance control accuracy and real-time performance. Therefore, there is an urgent need in the field for a high-efficiency feedforward control strategy that can accurately predict and compensate for path-dependent thermal effects. SUMMARY

[0004] In view of this, the present application provides a molten pool temperature feedforward control method that can predict thermal behavior and plan optimal laser power in advance based on processing path information, to achieve accurate and stable control of the molten pool temperature, thereby improving the quality consistency of laser additive manufactured parts.

[0005] To achieve the above purpose, the present application is implemented by the following technical solutions:

[0006] The present application provides a laser additive manufacturing molten pool temperature feedforward control method, comprising the following steps: step one: obtaining the surface temperature field after solidification of the Nth layer through a thermal imaging system as the initial state T0;

[0007] Step two: obtaining the scanning path of the N+1th layer from the CAM system, calculating the equivalent heat dissipation index of each point on the path, and generating an EHDI sequence;

[0008] Step three: inputting T0 and the EHDI sequence into a pre-trained LSTM prediction model to predict the dynamic change of the molten pool temperature;

[0009] Step four: based on the model prediction path integral control algorithm, taking the minimization of the cost function containing the temperature tracking error, power change and power size as the goal, and optimizing to generate the optimal laser power feedforward control sequence of the N+1th layer.

[0010] In the laser additive manufacturing molten pool temperature feedforward control method, as a preferred scheme, the calculation formula of the equivalent heat dissipation index in step two is:

[0011]

[0012] Wherein: V(p i ) is the volume ratio of the solid phase in the local hemisphere volume of the geometric heat dissipation component of the current machining point radius r, which can be approximately calculated after discretization and voxelization of the CAM model; T hist (p i , t i ) is the heat accumulation component; w g , w t are weight coefficients;

[0013] The geometric heat dissipation component is a physical quantity jointly acting by the distance from the current machining point to the part boundary, the density of the adjacent scanned track, and the local geometric features;

[0014] The heat accumulation component is the preheating effect of the adjacent area of the point in the nearest time window dominated by the scanning path.

[0015] In the laser additive manufacturing molten pool temperature feedforward control method, as a preferred scheme, the heat accumulation is calculated by the following formula:

[0016]

[0017] Wherein , is the initial temperature, η is the thermal conversion efficiency, P k is the input power of the kth time step, Δt k is the time interval of the kth time step.

[0018] In the laser additive manufacturing molten pool temperature feedforward control method, as a preferred scheme, the LSTM model in step three is obtained by training historical machining data, and the training data includes the synchronously collected laser power sequence, the molten pool temperature sequence, and the corresponding EHDI sequence.

[0019] In the laser additive manufacturing molten pool temperature feedforward control method, as a preferred scheme, the hyperparameters of the LSTM model are automatically optimized by using an optimization algorithm.

[0020] In the laser additive manufacturing molten pool temperature feedforward control method, as a preferred scheme, the cost function J in step four is:

[0021]

[0022] Wherein, Tt T is the predicted molten pool temperature at time t ref T is the target molten pool temperature k+N T is the predicted molten pool temperature at time t t w is the laser power at time t T w is the laser power at time t ΔP w is the laser power at time t F are the weight coefficients of control quality, control action cost and final control effect, respectively.

[0023] In the laser additive manufacturing molten pool temperature feedforward control method described above, as a preferred scheme, the implementation of the MPPI algorithm in step four includes:

[0024] S1, initialize parameters; based on material properties, initial equipment state and process parameters;

[0025] S2, temperature field addition; based on the temperature field at the end of the previous layer processing;

[0026] S3, parallel perturbation sampling; parallel sampling of multiple laser power perturbation sequences;

[0027] S4, LSTM forward prediction; using the LSTM model to forward simulate each power sequence and predict the temperature trajectory;

[0028] S5, cost evaluation and iterative optimization; evaluate the cost of each trajectory according to the cost function, and perform weighted average to update the control sequence, and after iterative convergence, obtain the optimal laser power feedforward control sequence.

[0029] In the laser additive manufacturing molten pool temperature feedforward control method described above, as a preferred scheme, the optimal laser power feedforward control sequence is calculated during the interlayer powder laying stage, and is executed synchronously with the scanning path during the layer scanning process.

[0030] In the laser additive manufacturing molten pool temperature feedforward control method described above, as a preferred scheme, the anti-winding control method is a disturbance observer-based anti-disturbance control method, and the disturbance observer is a state space model or transfer function used to quantitatively estimate the disturbance size of the system; assuming that the disturbance model is , then the system state model containing the disturbance component can be obtained; then, the optimal linear controller based on this model is designed using the least square adjustment method to realize anti-disturbance control.

[0031] In the laser additive manufacturing molten pool temperature feedforward control method, as a preferred scheme, a hardware device for running the control method is further included, and the hardware device includes a laser, an atmosphere control system, a material delivery system, an electrical system, a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the method of any one of claims 1-9 is implemented.

[0032] The application provides a laser additive manufacturing molten pool temperature feedforward control method, which has the following beneficial effects:

[0033] The application can "predict" thermal disturbance before scanning starts, generate an optimal power scheme, and realize accurate feedforward control of the molten pool temperature by quantifying path information into an EHDI sequence and accurately predicting thermal dynamics by using an LSTM model; the evolution trend of the molten pool temperature is predicted by the machining path, and the adjustment value of the laser power is given in advance, which can effectively reduce part defects (such as pores and cracks), improve microstructure uniformity, reduce residual stress, and ultimately improve the mechanical properties and dimensional accuracy of the formed parts;

[0034] The MPPI method does not depend on the differentiability of the controlled model and can effectively handle complex nonlinear and path-dependent systems such as laser additive manufacturing;

[0035] The MPPI algorithm is highly parallel and can be accelerated by GPU to complete the power planning of the entire layer within the short powder laying time between layers, meeting the real-time requirements of engineering. BRIEF DESCRIPTION OF DRAWINGS

[0036] Fig. 1 The application provides a laser additive manufacturing molten pool temperature feedforward control method, which has the following beneficial effects:

[0037] Fig. 2 The application provides a laser additive manufacturing molten pool temperature feedforward control method, which has the following beneficial effects:

[0038] Fig. 3 The application provides a laser additive manufacturing molten pool temperature feedforward control method, which has the following beneficial effects: DETAILED DESCRIPTION

[0039] To make the purpose, technical scheme and advantages of the embodiments of the application clearer, the technical scheme in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the application.

[0040] The exemplary embodiments of the application will be described in detail with reference to the following drawings.

[0041] The following describes exemplary embodiments of the present invention in conjunction with specific circumstances:

[0042] like Figs. 1 to 3 As shown, a laser additive manufacturing melt pool temperature feedforward control method includes the following steps: Step 1: Obtain the surface temperature field of the Nth layer after solidification using a thermal imaging system, as the initial state T0; The thermal imaging system refers to a system based on a near-infrared or short-wave infrared external array camera and an adapter lens to collect high-temperature thermal signals from the melt pool. It is usually mounted on the laser processing device in a coaxial or off-axis manner. After processing and synthesizing the collected melt pool thermal signals, the system can obtain a thermal distribution map of the formed layer, which is an optical tomographic image of the thermal input history of that layer. The thermal imaging system is the detection link in the MPPI control system of this invention, providing key additive manufacturing forming process status feedback for the control algorithm.

[0043] Step 2: Obtain the scan path of layer N+1 from the CAM system, calculate the equivalent heat dissipation index at each point on the path, and generate the EHDI sequence; specifically: the formula for calculating the equivalent heat dissipation index in Step 2 is:

[0044]

[0045] Where: V(p) i The geometric heat dissipation component is the volume ratio of the solid phase within the local hemispherical volume at the current processing point radius r. This ratio can be approximated by discretizing and voxelizing the CAM model; T hist (p i , t i ) represents the heat accumulation component; w g , w t These are the weighting coefficients;

[0046] The geometric heat dissipation component is a physical quantity resulting from the combined effects of the distance from the current machining point to the part boundary, the density of nearby scanned trajectories, and local geometric features.

[0047] The heat accumulation component is the preheating effect of the scan path in the vicinity of the point within the most recent time window.

[0048] Heat accumulation is calculated using the following formula:

[0049]

[0050] in , Where is the initial temperature (for laser metal additive manufacturing, the possible range of molten pool temperature is 0 to 4000°C), η is the heat conversion efficiency (its possible range is 0 to 1), and P... k Let Δt be the input power at the k-th time step (for laser metal additive manufacturing, the possible range of laser power is 0 to 200 kW).k the time interval for the kth time step (for laser metal additive manufacturing, At k may range from 0 to 1 s).

[0051] The LSTM model is trained by historical processing data, and the training data includes synchronously collected laser power sequence, molten pool temperature sequence and corresponding EHDI sequence. The hyperparameters of the LSTM model are automatically optimized by using an optimization algorithm.

[0052] Step three: input T0 and EHDI sequence into the pre-trained LSTM prediction model to predict the dynamic change of the molten pool temperature; the LSTM prediction model is a deep learning model for predicting time series data, and due to the time correlation of the molten pool temperature before and after the processing, the LSTM model can better represent the influence of process parameters and scanning path on the molten pool temperature. And using a pre-trained model only needs to perform model inference to get the expected result, which is an efficient deep learning model application method.

[0053] The LSTM model is trained by historical processing data, and the training data includes synchronously collected laser power sequence, molten pool temperature sequence and corresponding EHDI sequence. The hyperparameters of the LSTM model are automatically optimized by using an optimization algorithm. The so-called historical processing data here is an input-output data pair that can reflect the correlation between the process input parameters and the forming process state, including but not limited to laser power-molten pool temperature, scanning speed-molten pool size, laser power and scanning speed-molten pool size and molten pool temperature, etc.

[0054] Step four: based on the model prediction path integral control algorithm, the optimal laser power feedforward control sequence of N+1 layers is optimized to minimize the cost function including temperature tracking error, power change and power size.

[0055] The cost function J in step four is:

[0056]

[0057] where T t is the predicted molten pool temperature at time t, T ref is the target molten pool temperature, T k+N is the predicted molten pool temperature at time t, for laser metal additive manufacturing, the possible range of the molten pool temperature is 0 to 4000°C; P t is the laser power at time t, for laser metal additive manufacturing, the possible range of the laser power is 0 to 200kW; w T , w ΔP , w FThe weight coefficients of the control quality, the control action cost and the final control effect, respectively, and their value ranges are all 0 to 1.

[0058] The implementation of the MPPI algorithm in step four comprises:

[0059] S1, initialization of parameters; based on material properties, initial equipment state and process parameters;

[0060] S2, temperature field addition; based on the temperature field at the end of the previous layer processing;

[0061] S3, parallel perturbation sampling; parallel sampling of multiple laser power perturbation sequences;

[0062] S4, LSTM forward prediction; using the LSTM model to perform forward simulation on each power sequence to predict the temperature trajectory;

[0063] S5, cost evaluation and iterative optimization; evaluate the cost of each trajectory according to the cost function, and perform weighted average to update the control sequence, and after iterative convergence, the optimal laser power feedforward control sequence is obtained.

[0064] The optimal laser power feedforward control sequence is calculated during the interlayer powder laying stage and is executed synchronously with the scanning path during the layer scanning process.

[0065] The anti-winding control method is an anti-disturbance control method based on a disturbance observer, and the anti-disturbance observer is a state space model or transfer function used to quantitatively estimate the size of the disturbance to which the system is subjected; assuming that the disturbance model is , then the system state model containing the disturbance component can be obtained; then, the optimal linear controller based on this model is designed by using the least square adjustment method to realize anti-disturbance control.

[0066] In the laser additive manufacturing molten pool temperature feedforward control method described above, as a preferred scheme, a hardware device for running the control method is further included, and the hardware device includes a laser, an atmosphere control system, a material conveying system, an electrical system, a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the method according to any one of claims 1-9 is realized.

[0067] Embodiment two: a laser additive manufacturing molten pool temperature feedforward control method, comprising the following steps:

[0068] Step one: processing path information quantification (equivalent heat dissipation index sequence generation)

[0069] The laser scanning path of the current processing layer is obtained from the laser additive manufacturing plan file planned by the CAM software, which is composed of a series of coordinate points arranged in time sequence. For each point on the path, its equivalent heat dissipation index (EHDI) is calculated. The index is a comprehensive quantitative value, and the calculation formula is:

[0070]

[0071] wherein: G(p i ) is the geometric heat dissipation component, which is determined by the distance from the point to the part boundary, the density of the adjacent scanned trajectory, and the local geometric features (such as whether it is a cantilever structure); T hist (p i , t i ) is the heat accumulation component, which is determined by the scanning history (preheating effect) of the adjacent area of the point in the nearest time window; w g , w t are weight coefficients. Traverse the entire path to generate a time series EHDI t corresponding to the path points.

[0072] Step two: dynamic prediction model construction and training of the molten pool temperature

[0073] A long short-term memory network model LSTM is constructed as a dynamic predictor of the molten pool temperature. The LSTM model input is the current and historical time laser power P t and equivalent heat dissipation index EHDI t . The LSTM model output is the predicted molten pool temperature T t+1 at the next time. The LSTM model is trained using a training set containing multiple sets of (P t , EHDI t , T t ) aligned data, and the model prediction accuracy is improved through hyperparameter optimization (such as adaptive moment estimation, Adam).

[0074] Step three: optimal power planning based on model prediction path integral control (MPPI)

[0075] The trained LSTM model is used as the forward prediction model of MPPI. The cost function J used to evaluate the control effect is defined as:

[0076]

[0077] wherein, T t is the predicted molten pool temperature at time t, T ref is the target molten pool temperature, and T k+NTo predict the final melt pool temperature of the step size, P t Let w be the laser power at time t. T , w ΔP , w F These are the weighting coefficients for control quality, control action overhead, and final control effect, respectively. This function simultaneously considers temperature tracking error, laser power variation amplitude, and final control error.

[0078] Before the start of a new layer of processing, using the temperature field at the end of the previous layer as the initial state, and combining it with the EHDI sequence of the current layer, the MPPI optimization algorithm is run:

[0079] 1. Sampling: Generate a large number of candidate laser power perturbation sequences in parallel.

[0080] 2. Forward simulation: Use an LSTM model to predict the future molten pool temperature trajectory corresponding to each candidate power sequence.

[0081] 3. Evaluation and weighting: Calculate the cost of each trajectory based on the cost function and assign high weights to low-cost sequences.

[0082] 4. Iterative Update: The base power sequence is updated by weighted averaging. After multiple iterations, the process converges to obtain the optimal feedforward laser power curve P for the current layer. optimal .

[0083] Step 4: Disturbance immunity control of molten pool temperature based on disturbance observer

[0084] During the processing of this layer, the control system not only synchronously executes the optimal power curve and scan path, but also further maintains the stability of the molten pool temperature through an anti-winding control loop based on a disturbance observer. Considering the ability to suppress higher-order disturbances, a second-order disturbance model is assumed to be... Therefore, the designed disturbance observer can be written as: If the system is completely observable, then It can be arbitrarily configured and designed using the linear quadratic control (LQR) method. The anti-winding control output is obtained as The total control output is then u(t) = P. optimal (t)+ u dd (t), that is, the laser power at time t is the sum of the output of MPPI and the anti-interference controller.

[0085] In a typical LDED process, the steps of implementing this invention are as follows:

[0086] 1. After completing the Nth layer of powder application and before scanning the N+1th layer, the surface temperature field of the Nth layer after solidification is obtained by a thermal imaging system as the initial state T0.

[0087] 2. Read the scan path of the N+1th layer from the CAM system, calculate the EHDI sequence of the whole path according to the method described in step one of the present application.

[0088] 3. Input T0 and the EHDI sequence into the pre-trained LSTM prediction model.

[0089] 4. Start the MPPI optimizer, perform parallel sampling and iterative optimization on the GPU according to the process described in step three (for example, sample 10,000 sequences and iterate 50 times), to generate the optimal laser power curve for the N+1th layer.

[0090] 5. Load the power curve into the laser controller, and when the galvanometer or mechanical arm in the laser additive system starts scanning the N+1th layer, the laser power will be adjusted synchronously and accurately according to the planned curve.

[0091] Embodiment three: hyperparameter optimization algorithm based on adaptive cosine annealing learning rate planning method, including the following steps:

[0092] Step one: initialize learning rate planning parameters. Set the initial learning rate , the minimum learning rate , the basic number of epochs , the performance improvement threshold δ, the upper limit of the patience counter , and the initial number of epochs E0.

[0093] Step two: perform the learning rate warm-up phase. In the first E0 training epochs, the learning rate remains at the initial value .

[0094] Step three: after the initial E0 training epochs are completed, enter the adaptive cosine annealing phase. For the E t th training cycle (E t > E0), calculate the position of the current cycle in the cosine function: pos = E t - E0, then the current learning rate can be calculated by the cosine annealing formula:

[0095]

[0096] where is the number of training epochs planned for the current stage.

[0097] Step four: monitor model performance. After each training cycle, calculate the loss function value on the validation set, and compare it with the historical best loss . If , it is determined that the performance has improved, and , reset the patience counter p = 0, continue to perform step three until the condition of completing the training is reached; otherwise, the patience counter is incremented , if , early stopping is performed.

[0098] The condition of completing the training can generally be set as the loss function value being less than a given threshold, i.e. ; or completing a given number of rounds E F .

[0099] Through comparison experiments, after the method is used, the root mean square error (RMSE) of the molten pool temperature can be reduced by more than 60% compared with the constant power strategy, the porosity of the formed part is significantly reduced, and the microstructure is more uniform.

[0100] Finally, it should be noted that in this document, relational terms such as first and second and the like can merely be used to distinguish one entity or action from another, without necessarily requiring or implying any actual such relationship or order between or among entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises... a" does not, without more constraints, exclude the existence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0101] The above describes the principles and implementation manners of the present application by using specific examples, and the above example is only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manner and application range can be changed, and the above description should not be understood as limiting the present application.

Claims

1. A laser additive manufacturing melt pool temperature feedforward control method, characterized by, The method comprises the following steps: Step 1: Obtain the surface temperature field of the Nth layer after solidification by a thermal imaging system as the initial state T0; Step 2: Obtain the scanning path of the N+1th layer from the CAM system, calculate the equivalent heat dissipation index of each point on the path, and generate an EHDI sequence; Step 3: Input T0 and the EHDI sequence into a pre-trained LSTM prediction model to predict the dynamic change of the molten pool temperature; Step 4: Based on the model prediction path integral control algorithm, the optimal laser power feedforward control sequence of the N+1th layer is generated by minimizing the cost function containing the temperature tracking error, power change and power size as the target.

2. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein: The calculation formula of the equivalent heat dissipation index in step 2 is: wherein: V(p i ) is the volume ratio of the solid phase in the local hemisphere volume of the current machining point radius r of the geometric heat dissipation component, which can be approximately calculated after discretization and voxelization of the CAM model; T hist (p i , t i ) is the heat accumulation component; w g , w t are weight coefficients; The geometric heat dissipation component is a physical quantity jointly acted by the distance from the current processing point to the part boundary, the density of adjacent scanned tracks and local geometric features; The heat accumulation component is the preheating effect of the adjacent area within the nearest time window dominated by the scanning path.

3. The laser additive manufacturing melt pool temperature feedforward control method of claim 2, wherein: The heat accumulation is calculated by the following formula: wherein , T0is the initial temperature, η is the thermal conversion efficiency, Pkis the input power at the kth time step, and Δtkis the time interval at the kth time step.

4. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein: The LSTM model in step 3 is obtained by training historical processing data, and the training data includes the synchronous acquisition of laser power sequence, molten pool temperature sequence and corresponding EHDI sequence.

5. The laser additive manufacturing melt pool temperature feedforward control method of claim 4, wherein: The hyperparameters of the LSTM model are automatically optimized by an optimization algorithm.

6. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein, The cost function J in step 4 is: Wherein, T t is the predicted t moment molten pool temperature, T ref is the target molten pool temperature, T k+N is the predicted t moment molten pool temperature, P t is the laser power at t moment, w T , w ΔP , w F are weight coefficients of control quality, control action cost and final control effect respectively.

7. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein The implementation of the MPPI algorithm in step 4 includes: S1, initialize parameters; based on material properties, initial device state and process parameters; S2, temperature field addition; based on the temperature field at the end of the last layer processing; S3, parallel perturbation sampling; parallel sampling of multiple laser power perturbation sequences; S4, LSTM forward prediction; use the LSTM model to forward simulate each power sequence to predict the temperature trajectory; S5, cost evaluation and iterative optimization; evaluate the cost of each trajectory according to the cost function, and perform weighted average to update the control sequence, and after iterative convergence, the optimal laser power feedforward control sequence is obtained.

8. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein: The optimal laser power feedforward control sequence is calculated during the layer powder laying stage and is executed synchronously with the scanning path during the layer scanning process.

9. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein: The anti-winding control method is an anti-disturbance control method based on a disturbance observer, the disturbance observer is a state space model or a transfer function for quantitatively estimating the disturbance size of a system; assuming that a disturbance model is , a system state model containing a disturbance component can be obtained; then, an optimal linear controller based on the model is designed by using a least square adjustment method, so as to realize anti-disturbance control.

10. The laser additive manufacturing melt pool temperature feedforward control method of claim 1, wherein: Further comprising a hardware device for running the control method, the hardware device comprising a laser, an atmosphere control system, a material conveying system, an electrical system, a memory and a processor, the memory storing a computer program which, when executed by the processor, implements the method of any one of claims 1-9.

Images