A motor control method and system based on high-order filter

Abstract

The present invention provides a motor control method and system based on a high-order filter, which relates to the field of motor control technology. The method includes obtaining initial infrared camera sensor sequence data of the previous layer of 3D printing; removing high-frequency noise from the initial infrared camera sensor sequence data of the previous layer of 3D printing based on a high-order filter to obtain filtered infrared camera sensor sequence data of the previous layer of 3D printing; determining multiple first low-temperature point information and multiple first low thermal stress point information based on a thermal map and a thermal stress distribution map of the previous layer of 3D printing; generating a motor control scheme based on the 3D printing start path information and the 3D printing subsequent path information of the next layer, and performing motor control based on the motor control scheme. The method can quickly and accurately determine the optimal printing path for 3D printing to improve printing accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of motor control, and in particular to a motor control method and system based on a high-order filter. Background Art

[0002] In the manufacturing industry, 3D printing technology, with its unique method of constructing three-dimensional objects by stacking materials layer by layer, is gradually transforming traditional manufacturing processes. During 3D printing, the precision of the print head's motion control plays a critical role in print quality. As the print head builds the three-dimensional structure layer by layer, its path planning and temperature control directly impact the quality of the final product. In actual printing, existing technologies often cause defects such as warping, deformation, and poor interlayer bonding due to factors such as material thermal expansion and accumulated thermal stress. These issues not only affect printing success rates but also limit the application of 3D printing technology in high-precision manufacturing. Currently, most 3D printing systems rely on pre-defined print paths and fixed parameters. These systems lack in-depth analysis of thermal stress distribution, making it difficult to accurately identify potential high-risk areas such as high-temperature zones and areas of high stress concentration. Existing print path planning methods typically directly generate the print path based on a geometric model, failing to fully consider the temperature and stress conditions left over from the previous layer. This can lead to localized overheating or stress accumulation when printing the next layer, exacerbating structural defects.

[0003] Therefore, how to quickly and accurately determine the optimal printing path for 3D printing to improve printing accuracy is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem solved by the present invention is how to quickly and accurately determine the optimal printing path for 3D printing to improve printing accuracy.

[0005] According to a first aspect, the present invention provides a motor control method based on a high-order filter, comprising: obtaining initial infrared camera sensor sequence data of a previous layer of 3D printing; removing high-frequency noise from the initial infrared camera sensor sequence data of the previous layer of 3D printing based on a high-order filter to obtain filtered infrared camera sensor sequence data of the previous layer of 3D printing; determining a thermal map and a thermal stress distribution map of the previous layer of 3D printing using an infrared analysis model based on the filtered infrared camera sensor sequence data of the previous layer of 3D printing; determining a plurality of first low-temperature point information and a plurality of first low-temperature point information based on the thermal map and the thermal stress distribution map of the previous layer of 3D printing. first low thermal stress point information; determining multiple 3D printing starting point information of the next layer based on the multiple first low temperature point information and the multiple first low thermal stress point information and the next layer printing information; determining the 3D printing starting path information of the next layer based on the multiple 3D printing starting point information of the next layer; determining the 3D printing subsequent path information based on the 3D printing starting path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the next layer printing information; generating a motor control scheme based on the 3D printing starting path information of the next layer and the 3D printing subsequent path information, and performing motor control based on the motor control scheme.

[0006] In one possible implementation, the determining of the 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the printing information of the next layer includes: determining multiple preliminary 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the printing information of the next layer; and determining the 3D printing subsequent path information based on the multiple preliminary 3D printing subsequent path information.

[0007] In one possible implementation, determining the 3D printing subsequent path information based on the multiple preliminary 3D printing subsequent path information includes: constructing a path map, the path map including multiple preliminary path nodes and edges between the multiple preliminary path nodes, the node features of each preliminary path node including one preliminary 3D printing subsequent path information, the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the edges between the nodes are the printing time differences of the preliminary 3D printing subsequent paths; processing the path map based on a graph neural network to determine the 3D printing subsequent path information.

[0008] In a possible implementation, the infrared analysis model is a recurrent neural network model.

[0009] According to a second aspect, the present invention provides a motor control system based on a high-order filter, comprising: a data acquisition module for acquiring initial infrared camera sensor sequence data of a previous layer of 3D printing; a filtering processing module for removing high-frequency noise from the initial infrared camera sensor sequence data of the previous layer of 3D printing based on a high-order filter to obtain filtered infrared camera sensor sequence data of the previous layer of 3D printing; a thermal analysis module for determining a thermal map and a thermal stress distribution map of the previous layer of 3D printing using an infrared analysis model based on the filtered infrared camera sensor sequence data of the previous layer of 3D printing; a feature extraction module for determining a plurality of first low-temperature point information and a plurality of first low-temperature point information based on the thermal map and the thermal stress distribution map of the previous layer of 3D printing. a printing point determination module for determining multiple 3D printing starting point information of the next layer based on the multiple first low temperature point information and the multiple first low thermal stress point information and the next layer printing information; a path planning module for determining the 3D printing starting path information of the next layer based on the multiple 3D printing starting point information of the next layer; a path optimization module for determining the 3D printing subsequent path information based on the 3D printing starting path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the next layer printing information; a control execution module for generating a motor control scheme based on the 3D printing starting path information of the next layer and the 3D printing subsequent path information, and performing motor control based on the motor control scheme.

[0010] In one possible implementation, the path optimization module is further used to: determine multiple preliminary 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous layer 3D printing, and the next layer printing information; determine the 3D printing subsequent path information based on the multiple preliminary 3D printing subsequent path information.

[0011] In one possible implementation, the path optimization module is further used to: construct a path map, the path map including multiple preliminary path nodes and edges between the multiple preliminary path nodes, the node features of each preliminary path node including a preliminary 3D printing subsequent path information, the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the edges between nodes are the printing time differences of the preliminary 3D printing subsequent paths; and process the path map based on a graph neural network to determine the 3D printing subsequent path information.

[0012] In a possible implementation, the infrared analysis model is a recurrent neural network model.

[0013] According to a third aspect, an embodiment of the present invention provides an electronic device, comprising: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method comprising: obtaining initial infrared camera sensor sequence data of the previous layer of 3D printing; removing high-frequency noise from the initial infrared camera sensor sequence data of the previous layer of 3D printing based on a high-order filter to obtain filtered infrared camera sensor sequence data of the previous layer of 3D printing; based on the filtered infrared camera sensor sequence data of the previous layer of 3D printing, using an infrared analysis model to determine a thermal map and a thermal stress distribution map of the previous layer of 3D printing; based on the previous layer of 3D printing The thermal map and thermal stress distribution map of a 3D printed layer determine multiple first low-temperature point information and multiple first low thermal stress point information; based on the multiple first low-temperature point information and multiple first low thermal stress point information and the printing information of the next layer, multiple 3D printing starting point information of the next layer is determined; based on the multiple 3D printing starting point information of the next layer, the 3D printing starting path information of the next layer is determined; based on the 3D printing starting path information of the next layer, the thermal map and thermal stress distribution map of the 3D printed previous layer, and the printing information of the next layer, the 3D printing subsequent path information is determined; a motor control scheme is generated based on the 3D printing starting path information of the next layer and the 3D printing subsequent path information, and the motor control is performed based on the motor control scheme.

[0014] According to a fourth aspect, the present embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned motor control method based on a high-order filter, the method comprising: obtaining initial infrared camera sensor sequence data of a previous layer of 3D printing; removing high-frequency noise from the initial infrared camera sensor sequence data of the previous layer of 3D printing based on a high-order filter to obtain filtered infrared camera sensor sequence data of the previous layer of 3D printing; determining a thermal map and a thermal stress distribution map of the previous layer of 3D printing based on the filtered infrared camera sensor sequence data of the previous layer of 3D printing using an infrared analysis model; and determining a thermal map and a thermal stress distribution map of the previous layer of 3D printing based on the thermal map of the previous layer of 3D printing. The method comprises the following steps: determining a plurality of first low-temperature point information and a plurality of first low thermal stress point information based on a force map and a thermal stress distribution map; determining a plurality of 3D printing starting point information of a plurality of next layers based on the plurality of first low-temperature point information and the plurality of first low thermal stress point information and the next layer printing information; determining a plurality of 3D printing starting path information of a next layer based on the plurality of 3D printing starting point information of the next layer; determining a 3D printing subsequent path information based on the 3D printing starting path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the next layer printing information; generating a motor control scheme based on the 3D printing starting path information of the next layer and the 3D printing subsequent path information, and performing motor control based on the motor control scheme.

[0015] The present invention provides a motor control method and system based on a high-order filter, the method comprising: obtaining initial infrared camera sensor sequence data of a previous 3D print; removing high-frequency noise from the initial infrared camera sensor sequence data of the previous 3D print using a high-order filter to obtain filtered infrared camera sensor sequence data of the previous 3D print; determining a thermal map and a thermal stress distribution map of the previous 3D print using an infrared analysis model based on the filtered infrared camera sensor sequence data of the previous 3D print; determining a plurality of first low-temperature point information and a plurality of first low thermal stress point information based on the thermal map and the thermal stress distribution map of the previous 3D print; and determining a plurality of first low-temperature point information based on the plurality of first low thermal stress point information. The method comprises the following steps: determining the 3D printing starting point information of multiple next layers based on the first low temperature point information, multiple first low thermal stress point information, and the next layer printing information; determining the 3D printing starting path information of the next layer based on the multiple 3D printing starting point information of the next layer; determining the 3D printing subsequent path information based on the 3D printing starting path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the next layer printing information; generating a motor control scheme based on the 3D printing starting path information of the next layer and the 3D printing subsequent path information, and performing motor control based on the motor control scheme. The method can quickly and accurately determine the optimal printing path for 3D printing to improve printing accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic flow chart of a motor control method based on a high-order filter provided by an embodiment of the present invention;

[0017] Figure 2 is a schematic diagram of a 3D printing device in an embodiment of the present invention;

[0018] Figure 3 A schematic diagram of a process for determining subsequent path information for 3D printing provided by an embodiment of the present invention;

[0019] Figure 4 A schematic diagram of a process for determining 3D printing subsequent path information based on multiple pieces of preliminary 3D printing subsequent path information provided by an embodiment of the present invention;

[0020] Figure 5 A schematic diagram of a motor control system based on a high-order filter provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0021] The present invention will be further described in detail below by means of specific embodiments in conjunction with the accompanying drawings. Similar elements in different embodiments are numbered with associated similar elements. In the following embodiments, many detailed descriptions are provided to enable the present invention to be better understood. However, those skilled in the art will readily appreciate that some of the features may be omitted under different circumstances, or may be replaced by other elements, materials, or methods. In some cases, some operations related to the present invention are not shown or described in the specification. This is to avoid the core of the present invention being overwhelmed by excessive descriptions, and for those skilled in the art, it is not necessary to describe these related operations in detail. They can fully understand the related operations based on the description in the specification and the general technical knowledge in the art.

[0022] In an embodiment of the present invention, there is provided Figure 1 A motor control method based on a high-order filter is shown, and the motor control method based on a high-order filter includes steps S1 to S8:

[0023] Step S1, obtaining the initial infrared camera sensor sequence data of the previous layer of 3D printing.

[0024] 3D printing is an additive manufacturing technology that builds three-dimensional objects by stacking materials layer by layer. After each layer is printed, the print head moves to the next position to lay down the next layer of material until the entire object is constructed. The 3D printing process is based on a CAD digital model file and uses a 3D printing device to transform the design into a physical object. Figure 2Schematic diagram of a 3D printing device according to an embodiment of the present invention. In the 3D printing device, a motor is used to precisely drive the print head to move along a predetermined path to ensure that each layer of material is accurately laid.

[0025] The initial infrared camera sensor sequence data for the previous 3D printing layer is a continuous sequence of sensor data acquired through real-time infrared camera monitoring of the previous layer's 3D printing process. This initial infrared camera sensor sequence data records the dynamic changes in temperature distribution at all locations on the surface of the previous layer during printing, including pixel-level temperature values ​​at each time point.

[0026] Step S2: removing high-frequency noise from the initial infrared camera sensor sequence data of the previous 3D printing layer using a high-order filter to obtain filtered infrared camera sensor sequence data of the previous 3D printing layer.

[0027] A high-order filter is a signal processing tool that can perform multi-order differential or integral processing on a signal. High-order filters can be used to eliminate noise interference from a signal. By setting specific frequency response characteristics, high-order filters can selectively filter different frequency components in the input signal. These high-order characteristics enable a more precise separation of the signal's useful components from the noise. In 3D printing sensor data processing, high-order filters can effectively suppress high-frequency noise while retaining the low-frequency useful information in the signal that reflects temperature changes.

[0028] The filtered infrared camera sensor sequence data for the previous 3D print layer is obtained by processing the initial infrared camera sensor sequence data through a high-order filter. This filtered infrared camera sensor sequence data removes high-frequency noise caused by sensor jitter and environmental interference, thereby more realistically reflecting the temperature radiation patterns of the previous 3D print layer.

[0029] The initial infrared camera sensor sequence data from the previous 3D print contains both the temperature signal of the printed area and high-frequency noise. This high-frequency noise can interfere with the interpretation of actual temperature changes. The high-order filter, through its internal multi-order filtering structure, performs frequency analysis on the initial infrared camera sensor sequence data and identifies the high-frequency components of the noise. The high-order filter then attenuates the high-frequency noise according to preset filtering parameters, while preserving the low-frequency, significant signal reflecting temperature changes. After filtering by the high-order filter, the noise content in the filtered infrared camera sensor sequence data is significantly reduced.

[0030] Step S3, based on the filtered infrared camera sensor sequence data of the previous 3D print, using an infrared analysis model to determine the thermal map and thermal stress distribution map of the previous 3D print.

[0031] In some embodiments, the infrared analysis model is a recurrent neural network model. The input of the infrared analysis model is the filtered infrared camera sensor sequence data of the previous layer of 3D printing, and the output of the infrared analysis model is the thermal map and thermal stress distribution map of the previous layer of 3D printing.

[0032] Recurrent neural networks (RNNs) include recurrent neural networks (RNNs), which are deep learning models capable of processing time series data. They can transmit historical information through hidden states. At each computational step, RNNs use the current input and the previous hidden state to generate a new hidden state, capturing temporal dependencies in sequential data.

[0033] The thermal map of the previous 3D print, generated using an infrared analysis model, visually displays the temperature distribution of the previous print area at the time of printing. Different colors are used to indicate the temperature at different locations within the print area. Color changes correspond to temperature differences and clearly show the distribution of high and low temperature areas within the print area.

[0034] The thermal stress distribution map for the previous 3D print is generated using an infrared analysis model, reflecting the magnitude and distribution of thermal stress in the previous print area at the time of printing. The value at each coordinate point in the thermal stress distribution map for the previous 3D print represents the stress intensity experienced by the material at that location.

[0035] The filtered infrared camera sensor sequence data from the previous 3D print contains information about the temperature radiation of the printed area over time. This data exhibits distinct temporal characteristics, meaning that the temperature state at each moment is correlated with the previous one. A recurrent neural network model utilizes the hidden state memory function within its recurrent structure to process this sequence data moment by moment. It can extract the dynamic characteristics of temperature changes by learning the correlation patterns between temperature data at different time points. After inputting the filtered infrared camera sensor sequence data into the network, the model transforms the temporal temperature information into spatial distribution features through nonlinear transformations across multiple layers of neurons. The model maps the temperature radiation intensity at different locations to corresponding color values ​​and generates a two-dimensional thermal map of the temperature distribution. Furthermore, the model incorporates implicit patterns such as the material's thermal expansion coefficient and temperature gradient to convert the temperature distribution data into thermal stress values, which are then converted into a thermal stress distribution map through spatial mapping.

[0036] Step S4: determining a plurality of first low-temperature point information and a plurality of first low-thermal-stress point information based on the thermal map and thermal stress distribution map of the previous 3D printing layer.

[0037] In some embodiments, a low-temperature analysis model can be used to determine information about a plurality of first low-temperature points and information about a plurality of first low-thermal-stress points. The low-temperature analysis model is a convolutional neural network model, the input of which is the thermal map and thermal stress distribution map of the previous 3D print layer, and the output of which is information about a plurality of first low-temperature points and information about a plurality of first low-thermal-stress points.

[0038] Convolutional neural network models include convolutional neural networks (CNNs), which are deep learning models that extract local features through convolutional layers. The convolutional layers of a CNN use sliding windows (filters) to locally perceive input data and gradually extract features from low-level to high-level layers, thereby capturing the spatial local correlations of the data. In addition to convolutional layers, the core components of a CNN include pooling layers and fully connected layers. Pooling layers reduce feature dimensionality to retain key features, while fully connected layers integrate the extracted features and output the results.

[0039] The first low-temperature point information is output by the low-temperature analysis model and reflects the location with the lowest temperature in the previous printing area. Each first low-temperature point information includes the specific location coordinates, temperature value, and local temperature gradient characteristics of the location.

[0040] The first low thermal stress point information is output by the low temperature analysis model and reflects the location with the lowest thermal stress in the previous printing area. Each first low thermal stress point information includes the location coordinates, stress value, and local stress gradient characteristics of the low thermal stress point.

[0041] The thermal map and thermal stress distribution map of the previous 3D print layer contain the spatial distribution characteristics of the surface temperature and stress of the previous print layer upon completion. These distribution characteristics provide a direct data basis for the model to identify low-temperature and low-thermal stress points. The initial convolutional layer of the convolutional neural network scans the thermal map pixel by pixel using a 3×3 convolution kernel. This kernel is optimized to be sensitive to temperature differences. When it detects a local area with a temperature significantly lower than the surrounding area, it generates a high response value and marks a potential low-temperature point. The intermediate convolutional layer expands the receptive field to aggregate these low-temperature point features into continuous low-temperature regions, while capturing key features such as area size and boundary shape to avoid misjudgment caused by local noise. The final decision layer, taking into account the characteristics of the printing material (such as the safe temperature threshold of 60°C for PLA materials), filters the low-temperature regions after aggregation. It then retains all coordinates with temperatures below the threshold and labels the low-temperature region to which each point belongs. This generates the first low-temperature point information containing specific location coordinates, temperature values, and local temperature gradient characteristics. When processing thermal stress distribution maps, the model uses multi-scale convolution kernels (5×5 and 7×7 kernels of varying sizes) to scan the maps, identifying flat regions with low rates of stress change. These regions exhibit stable stress distribution and are less susceptible to deformation due to the temperature overlay of subsequent printing. The convolutional neural network's intermediate layers further extract features related to the stress values ​​and distribution ranges in these flat regions. The final decision layer then selects locations within these flat regions where stress values ​​fall below a safety threshold based on the material's yield strength threshold. Each selected point is annotated with its specific location coordinates, stress value, and local stress gradient characteristics, generating information on multiple points of first-order low thermal stress.

[0042] Step S5 , determining a plurality of 3D printing start point information of the next layer based on the plurality of first low temperature point information, the plurality of first low thermal stress point information, and the next layer printing information.

[0043] The next layer of printing information is the process guidance information obtained by reading the CAD digital model file of the 3D printing model and the preset process parameter configuration. The next layer of printing information includes the spatial range data, material property indicators, and structural feature annotations of the next layer of printing.

[0044] Spatial range data is the two-dimensional plane coordinate definition of the current layer of 3D printing. Spatial range data can be used to clarify the printing area boundary of the layer in the XY plane.

[0045] In 3D printing, each layer is printed flat at a fixed Z-axis height. The spatial range data defines the planar area where the material needs to be deposited in the current layer through the extreme values ​​of the X-axis and Y-axis coordinates.

[0046] Material property indicators include thermal and mechanical performance data such as decomposition temperature and yield strength of the materials used.

[0047] Structural feature annotation is a targeted description of the structural characteristics of the next layer, including key structural details such as complex geometric areas, thin walls or fragile parts that are easily deformed in the model cross-section.

[0048] In some embodiments, information about the 3D printing starting points of multiple next layers can be determined using an initial positioning model. The initial positioning model is a Transformer model. The inputs to the initial positioning model are the information about the multiple first low-temperature points, the information about the multiple first low-thermal-stress points, and the next layer printing information. The output of the initial positioning model is information about the 3D printing starting points of multiple next layers.

[0049] The Transformer model is a deep learning architecture based on the self-attention mechanism. Using a multi-head attention mechanism, the Transformer model can process correlations between input elements in parallel and effectively model interactions between complex parameters. The Transformer model consists of a stack of encoders. Each encoder layer contains a self-attention sublayer and a feedforward neural network sublayer. The self-attention sublayer calculates correlation weights between input elements, allowing each element to focus on global information. The feedforward sublayer performs nonlinear feature transformations.

[0050] The 3D printing starting point information for the next layers is a set of candidate printing starting position information that meets low-temperature and low-stress conditions, output by the initial positioning model. Each starting point information includes the starting position coordinates, temperature safety factor, and stress safety margin.

[0051] The temperature safety factor is the ratio of the material decomposition temperature to the current temperature at the starting point of 3D printing.

[0052] The stress safety margin is the difference between the material strength and the current stress at the starting point of 3D printing. The stress safety margin represents the residual capacity of the starting point to resist deformation.

[0053] The information on multiple first low-temperature points and multiple first low-thermal-stress points reflects safe locations with low temperature and low stress within the previous layer's print area. The next-layer print information includes key constraints such as the next layer's print range and material properties. The Transformer model's encoder encodes this input information, performing self-attention calculations on the position and value of the first low-temperature and low-thermal-stress points. It then captures correlations between different points, such as positional proximity and parameter similarity, while incorporating the next-layer print information as a constraint feature into the encoding process, thereby establishing a connection between the upper and lower layers. Based on the encoder output features and the starting position requirements for the next layer, the decoder uses an attention mechanism to focus on points that meet the requirements of low temperature, low stress, and within the next layer's print range. The model evaluates candidate points by learning the optimal starting point selection rules from the print data, and then selects multiple print start point information that both avoids printing defects caused by high temperature and high stress and is compatible with the next layer's print layout.

[0054] Step S6: determining the 3D printing start path information of the next layer based on the multiple 3D printing start point information of the next layers.

[0055] In some embodiments, a starting path generation model can be used to determine the 3D printing starting path information for the next layer. The starting path generation model is a deep neural network model. The input of the starting path generation model is the 3D printing starting point information of the multiple next layers, and the output of the starting path generation model is the 3D printing starting path information for the next layer.

[0056] Deep neural network models include deep neural networks (DNNs), which are composed of multiple hidden layers. DNNs extract and map features from input data through multi-layer nonlinear transformations. Each layer of neurons receives the output of the previous layer and performs weighted calculations and activation function processing. Their deep structure enables them to learn complex nonlinear relationships and high-order features in the data. In complex decision-making tasks, DNNs can achieve precise mapping from input features to output results through data training.

[0057] The next layer's 3D printing start path information is generated by the start path generation model, efficiently connecting all specified start points for the next layer's printing. This information includes the direction of the start path, the length of each path segment, and the coordinates of key locations along the path.

[0058] The information from multiple 3D printing starting points for the next layer provides candidate starting locations for the next printing layer. The location distribution and parameter characteristics of these points form the basis for path planning. A deep neural network model takes as input the coordinates, temperature, stress, and other characteristics of the starting points. The first layer of neurons then extracts location-correlation features of the starting points (such as distance between points and distribution density). Intermediate layers further process these features and learn the mapping patterns between starting points and paths from past printings, such as how to connect multiple starting points to minimize print head drift and reduce temperature accumulation. Deeper layers of neurons then fuse these features to construct a global planning feature for the starting path. Through multiple layers of nonlinear transformations, the model evaluates the path feasibility of different starting point combinations, selecting a path solution that ensures smooth print head movement from the starting point and avoids overlapping high-temperature and high-stress areas. Ultimately, the model outputs the starting path information for the next layer, including details such as the path direction and key coordinates.

[0059] Step S7: determining subsequent 3D printing path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous 3D printing layer, and the next layer printing information.

[0060] In some embodiments, Figure 3 A schematic diagram of a process for determining subsequent path information for 3D printing provided by an embodiment of the present invention, wherein determining subsequent path information for 3D printing includes steps S21 to S22:

[0061] Step S21 , determining a plurality of preliminary selected 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous layer 3D printing, and the next layer printing information.

[0062] In some embodiments, determining a plurality of preliminary selected 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous layer 3D printing, and the next layer printing information includes steps S31 to S33:

[0063] Step S31 : determining a plurality of low-temperature points and a plurality of low-stress points close to the starting path based on the 3D printing starting path information of the next layer, the thermal map and the thermal stress distribution map of the 3D printing of the previous layer.

[0064] In some embodiments, a convolutional neural network can be used to determine multiple low-temperature points and multiple low-stress points close to the starting path.

[0065] The multiple low-temperature points near the starting path are a collection of specific locations with lower temperatures located around the starting path, determined using a convolutional neural network. Each low-temperature point near the starting path includes its location coordinates, temperature value, local temperature gradient characteristics, and distance from the starting path.

[0066] Multiple low-stress points near the starting path are a collection of specific locations around the starting path with low stress, determined using a convolutional neural network. Each low-stress point near the starting path includes its location coordinates, stress value, local stress gradient characteristics, and distance from the starting path.

[0067] The convolutional layer of a convolutional neural network uses a sliding window to scan the area surrounding the starting path in the thermal map. The convolution kernel, which is sensitive to temperature differences, identifies local points with significantly lower temperatures than the surrounding areas. After filtering out noise through the pooling layer, it identifies multiple low-temperature points near the starting path. The convolutional neural network also processes the thermal stress distribution map, detecting low-stress areas and selecting multiple low-stress points near the starting path.

[0068] Step S32, based on the multiple low-temperature points close to the starting path, the multiple low-stress points close to the starting path, and the next layer of printing information, determine multiple safety point-associated subsequent path segments, multiple temperature adaptation extension directions, multiple stress adaptation extension directions, and multiple structural obstacle avoidance boundary information.

[0069] In some embodiments, a deep neural network can be used to determine multiple safety point-associated subsequent path segments, multiple temperature adaptation extension directions, multiple stress adaptation extension directions, and multiple structural obstacle avoidance boundary information.

[0070] The subsequent path segments associated with safety points are feasible subsequent path segments output by the deep neural network. Each path segment contains the starting point coordinates, end point coordinates, path length, and the temperature and stress mean of the area through which the path passes.

[0071] The temperature adaptation extension direction is determined by a deep neural network to extend a path starting from a low temperature point with a gentle temperature change. The temperature adaptation extension direction includes the direction angle, the temperature change rate threshold, and the maximum extension distance in that direction.

[0072] The stress-adaptive extension direction is determined by a deep neural network to be the extension direction of a path starting from a low-stress point with a gentle stress change. The stress-adaptive extension direction includes the direction angle, the stress change rate threshold, and the maximum extension distance in that direction.

[0073] Structural obstacle avoidance boundary information is the information about the next layer of fragile structural regions that the subsequent path needs to avoid, as determined by a deep neural network. This information includes vertex coordinates, region type (e.g., thin-walled regions, complex geometric regions), and obstacle avoidance safety distance.

[0074] Through multi-layer processing, the hidden layers of the deep neural network extract the spatial correlation between low-temperature and low-stress points in the input data, calculate the distance, direction, temperature, and stress difference between the two points, and generate a set of candidate subsequent path segments associated with safe points. The model then predicts the extension direction with gentle temperature changes based on the gradient characteristics of the low-temperature points. The temperature-adaptive extension direction parameters are determined based on the material properties. Furthermore, the stress-adaptive extension direction parameters are determined based on the gradient characteristics of the low-stress points. The model can identify and annotate vulnerable structures based on structural features. Furthermore, through boundary detection, it generates an initial boundary and optimizes the safe obstacle avoidance distance, forming structural obstacle avoidance boundary information.

[0075] Step S33: determining a plurality of preliminary selected 3D printing subsequent path information based on the plurality of safety point-associated subsequent path segments, the plurality of temperature adaptation extension directions, the plurality of stress adaptation extension directions, the plurality of structural obstacle avoidance boundary information, and the next layer printing information.

[0076] In some embodiments, a plurality of preliminary selected 3D printing subsequent path information may be determined by a deep neural network.

[0077] Multiple preliminary 3D printing subsequent path information is a collection of candidate trajectory planning information output by the deep neural network, which is used to carry on the starting path of the next layer of 3D printing. Each preliminary 3D printing subsequent path information includes the extended trajectory direction, the length of each path segment, the connection coordinates, the turning point coordinates, and the coordinates of the area covered by the path.

[0078] The deep neural network receives as input multiple subsequent path segments associated with safety points, multiple temperature-adaptive extension directions, multiple stress-adaptive extension directions, multiple structural obstacle avoidance boundary information, and the next layer of printing information. During processing, the input layer of the deep neural network normalizes these multidimensional features. The first hidden layer then filters the subsequent path segments associated with safety points based on the temperature-adaptive extension directions and the stress-adaptive extension directions, retaining those whose angles with the safety directions are within a preset threshold. The second layer, incorporating the structural obstacle avoidance boundary information, uses a path collision detection algorithm to eliminate path segments that cross the obstacle avoidance area and smooths path segments near the boundary to ensure that the obstacle avoidance safety distance is met. The third layer, based on the spatial range data in the next layer of printing information, uses an area coverage algorithm to combine the filtered path segments into candidate paths that fully cover the printing area. Subsequent hidden layers evaluate each candidate path through a reinforcement learning mechanism, calculating metrics such as total path length, accumulated temperature, and stress variation, thereby generating multiple preliminary 3D printing subsequent path information that meet safety constraints and coverage requirements. The output layer can then sort these preliminary paths, comprehensively consider factors such as path efficiency, temperature control, stress distribution, and finally output multiple differentiated preliminary 3D printing subsequent path information.

[0079] Step S22 : determining 3D printing subsequent path information based on the plurality of pieces of preliminary selected 3D printing subsequent path information.

[0080] In some embodiments, Figure 4 A schematic diagram of a process for determining 3D printing subsequent path information based on multiple pieces of preliminary 3D printing subsequent path information is provided in an embodiment of the present invention. The process for determining 3D printing subsequent path information based on multiple pieces of preliminary 3D printing subsequent path information includes steps S41 to S42:

[0081] Step S41: construct a path map, wherein the path map includes multiple preliminary path nodes and edges between the multiple preliminary path nodes. The node features of each preliminary path node include a preliminary 3D printing subsequent path information, the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the edges between the nodes are the printing time differences of the preliminary 3D printing subsequent paths.

[0082] The path graph is composed of two parts: nodes and edges. It can be used to reflect the relationship between the initial 3D printing path information. The path graph contains multiple initial path nodes and the edges between them. The node characteristics of each initial path node include the information of the initial 3D printing path, the starting path information of the next layer of 3D printing, and the thermal map and thermal stress distribution map of the previous layer of 3D printing. The edges between nodes are the printing time differences of different initial 3D printing paths.

[0083] By constructing a path graph, multiple discrete primary 3D printing subsequent path information can be abstracted into a structured topological relationship model.

[0084] Step S42: Process the path graph based on a graph neural network to determine subsequent 3D printing path information.

[0085] A graph neural network (GNN) is a deep learning model capable of processing graph-structured data. It can update node features by aggregating information from neighboring nodes. In each iteration, the GNN transmits features from neighboring nodes to the target node through message passing and integrates this information using an aggregation function to generate updated node features. The input of the GNN is the path graph, and the output is information about the subsequent 3D printing path.

[0086] 3D printing subsequent path information is the complete continuous printing trajectory planning information that is output by the graph neural network and continues the path of the next layer of 3D printing. 3D printing subsequent path information includes the direction of the extended trajectory, the geometric parameters of each path segment, and the coordinates of the area covered by the path.

[0087] The extended trajectory direction is the path direction extending from the end point of the starting path and the overall trajectory trend.

[0088] The geometric parameters of each path segment include the length of the extension segment, the connection coordinates of adjacent path segments, and the specific position coordinates of the turning points.

[0089] The coordinates of the area covered by the path are the coordinate set of the structure area within the next layer printing range that needs to be covered by the subsequent path.

[0090] Constructing a path graph is fundamental to a graph neural network's ability to accurately determine subsequent 3D printing paths. The design of nodes, node features, and edges within the path graph directly influences the graph neural network's ability to understand and plan the printing path. Nodes provide spatial anchors, node features provide attribute constraints, and edges provide relationships. Together, the path graph forms the complete data structure for graph neural networks to handle path planning.

[0091] The multiple preliminary path nodes in the path graph are independent carriers of each preliminary 3D printing subsequent path, and each node corresponds to a preliminary 3D printing subsequent path. These nodes provide clear analysis units for the graph neural network. The node features include information about a preliminary 3D printing subsequent path, information about the starting path for 3D printing of the next layer, and a thermal map and thermal stress distribution map of the previous layer of 3D printing. These features fully provide the specific content of each preliminary path, the starting background for generating the path, and the temperature and stress constraints that need to be followed. The edges define the printing time difference of different preliminary 3D printing subsequent paths. The edges are used to connect different preliminary path nodes and to establish efficiency associations between different preliminary paths. The time efficiency differences between the various preliminary paths can be intuitively presented through the edges. The graph neural network can identify the preliminary path with better printing time by comparing the values ​​of the edges.

[0092] The graph neural network performs initial feature extraction on each node in the path graph. Based on the node's features, it analyzes the path's extended trajectory, geometric parameters, and coverage area. It then evaluates the connectivity between the initial path and the next-level 3D printing start path, combining it with the next-level 3D printing start path information. The graph neural network then uses the thermal map and thermal stress distribution map of the previous 3D print layer to determine whether the path meets temperature and stress constraints. Through a message-passing mechanism, each node aggregates information transmitted by adjacent nodes via edges. Edges carry the printing time differences between different initial 3D printing subsequent paths. Edges allow nodes to perceive the time efficiency differences between their own path and other paths. During iterative updates, the graph neural network continuously integrates the node's own path details, background constraint features, and time efficiency information from adjacent nodes, gradually strengthening its assessment of the feasibility of each initial path, including smooth connectivity, constraint compliance, and efficiency. Ultimately, through a global analysis of all node features, the graph neural network can screen out a preliminary path that is precisely connected to the starting path of the next layer, meets the temperature stress constraints of the previous layer, and has the optimal printing time efficiency, so as to determine it as the subsequent path information for 3D printing.

[0093] Step S8: generating a motor control scheme based on the 3D printing start path information of the next layer and the 3D printing subsequent path information, and performing motor control based on the motor control scheme.

[0094] In some embodiments, a scheme determination model can be used to generate a motor control scheme, wherein the scheme determination model is a deep neural network model, the input of the scheme determination model is the 3D printing start path information of the next layer and the 3D printing subsequent path information, and the output of the scheme determination model is the motor control scheme.

[0095] A motor control scheme is a command scheme generated by a solution determination model to control the operation of the motor in a 3D printing device. The motor control scheme includes control parameters such as the motor's speed, direction, and start and stop times.

[0096] The next layer's 3D printing start path information and subsequent 3D printing path information define the trajectory and path details of the print head in the 3D printing device. The model takes path features such as coordinate sequence, path length, and curvature changes as input. Through nonlinear transformations across multiple layers of neurons, it learns the mapping between path features and motor control parameters. The model also optimizes control parameters based on the characteristics of the printing material (such as the motor's response speed relative to the melting temperature) to ensure that changes in motor speed and direction match the path characteristics. Ultimately, it outputs a motor control solution containing specific control instructions for precise control of the 3D printed motor.

[0097] Based on the same inventive concept, Figure 5 A schematic diagram of a motor control system based on a high-order filter provided by an embodiment of the present invention, wherein the motor control system based on the high-order filter includes:

[0098] The data acquisition module 51 is used to obtain the initial infrared camera sensor sequence data of the previous layer of 3D printing;

[0099] A filtering processing module 52 is configured to remove high-frequency noise from the initial infrared camera sensor sequence data of the previous 3D printing layer using a high-order filter to obtain filtered infrared camera sensor sequence data of the previous 3D printing layer;

[0100] A thermal analysis module 53 is configured to determine a thermal map and a thermal stress distribution map of the previous 3D print layer using an infrared analysis model based on the filtered infrared camera sensor sequence data of the previous 3D print layer;

[0101] A feature extraction module 54 is configured to determine a plurality of first low-temperature point information and a plurality of first low-thermal-stress point information based on the thermal map and thermal stress distribution map of the previous 3D printing layer;

[0102] A printing point determination module 55 is configured to determine a plurality of 3D printing start point information of the next layer based on the plurality of first low temperature point information, the plurality of first low thermal stress point information, and the next layer printing information;

[0103] A path planning module 56 is configured to determine 3D printing start path information of a next layer based on the 3D printing start point information of the multiple next layers;

[0104] A path optimization module 57 is configured to determine subsequent 3D printing path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous 3D printing layer, and the next layer printing information;

[0105] The control execution module 58 is configured to generate a motor control scheme based on the 3D printing start path information of the next layer and the 3D printing subsequent path information, and perform motor control based on the motor control scheme.

[0106] It should be noted that, in order to simplify the presentation of this specification and facilitate understanding of one or more embodiments of the invention, the foregoing descriptions of the embodiments of this specification sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this disclosure method does not imply that the subject matter of this specification requires more features than those recited in the claims. In fact, an embodiment may have fewer features than all of the features of a single disclosed embodiment.

[0107] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. A motor control method based on a high-order filter, characterized in that: include: Get the initial infrared camera sensor sequence data of the previous layer of 3D printing; Removing high-frequency noise from the initial infrared camera sensor sequence data of the previous 3D printing layer using a high-order filter to obtain filtered infrared camera sensor sequence data of the previous 3D printing layer; Determining a thermal map and a thermal stress distribution map of the previous 3D print using an infrared analysis model based on filtered infrared camera sensor sequence data of the previous 3D print; Determining a plurality of first low-temperature point information and a plurality of first low-thermal-stress point information based on the thermal map and thermal stress distribution map of the previous 3D print; Determining 3D printing start point information of multiple next layers based on the multiple first low temperature point information, the multiple first low thermal stress point information, and the next layer printing information; Determining 3D printing start path information for the next layer based on the 3D printing start point information for the multiple next layers; Determining subsequent 3D printing path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous 3D printing layer, and the next layer printing information; A motor control scheme is generated based on the 3D printing start path information of the next layer and the 3D printing subsequent path information, and motor control is performed based on the motor control scheme.

2. The motor control method based on a high-order filter according to claim 1, wherein: The determining of the subsequent path information for 3D printing based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the printing information of the next layer includes: Determining a plurality of preliminary selected 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the printing information of the next layer; 3D printing subsequent path information is determined based on the multiple pieces of preliminary selected 3D printing subsequent path information.

3. The motor control method based on a high-order filter according to claim 2, wherein: The determining of the 3D printing subsequent path information based on the plurality of pieces of preliminary selected 3D printing subsequent path information includes: Constructing a path map, the path map comprising a plurality of preliminary path nodes and edges between the plurality of preliminary path nodes, wherein node features of each preliminary path node include information about a preliminary 3D printing subsequent path, information about the start path of 3D printing of the next layer, a thermal map and a thermal stress distribution map of the 3D printing of the previous layer, and edges between nodes representing printing time differences of the preliminary 3D printing subsequent paths; The path map is processed based on a graph neural network to determine subsequent path information for 3D printing.

4. The motor control method based on a high-order filter according to claim 1, wherein: The infrared analysis model is a recurrent neural network model.

5. A motor control system based on a high-order filter, characterized in that: include: The data acquisition module is used to obtain the initial infrared camera sensor sequence data of the previous layer of 3D printing; A filtering processing module, configured to remove high-frequency noise from the initial infrared camera sensor sequence data of the previous 3D printing layer using a high-order filter to obtain filtered infrared camera sensor sequence data of the previous 3D printing layer; a thermal analysis module for determining a thermal map and a thermal stress distribution map of the previous 3D print layer using an infrared analysis model based on filtered infrared camera sensor sequence data of the previous 3D print layer; A feature extraction module, configured to determine a plurality of first low-temperature point information and a plurality of first low-thermal-stress point information based on the thermal map and thermal stress distribution map of the previous 3D print layer; a printing point determination module, configured to determine a plurality of 3D printing start point information of a next layer based on the plurality of first low temperature point information, the plurality of first low thermal stress point information, and the next layer printing information; A path planning module, configured to determine 3D printing start path information for a next layer based on the 3D printing start point information for the multiple next layers; a path optimization module, configured to determine subsequent 3D printing path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the previous 3D printing layer, and the next layer printing information; The control execution module is used to generate a motor control scheme based on the 3D printing start path information of the next layer and the 3D printing subsequent path information, and perform motor control based on the motor control scheme.

6. The motor control system based on a high-order filter according to claim 5, characterized in that: The path optimization module is also used for: Determining a plurality of preliminary selected 3D printing subsequent path information based on the 3D printing start path information of the next layer, the thermal map and thermal stress distribution map of the 3D printing of the previous layer, and the printing information of the next layer; 3D printing subsequent path information is determined based on the multiple pieces of preliminary selected 3D printing subsequent path information.

7. The motor control system based on a high-order filter according to claim 6, characterized in that: The path optimization module is also used for: Constructing a path map, the path map comprising a plurality of preliminary path nodes and edges between the plurality of preliminary path nodes, wherein node features of each preliminary path node include information about a preliminary 3D printing subsequent path, information about the start path of 3D printing of the next layer, a thermal map and a thermal stress distribution map of the 3D printing of the previous layer, and edges between nodes representing printing time differences of the preliminary 3D printing subsequent paths; The path map is processed based on a graph neural network to determine subsequent path information for 3D printing.

8. The motor control system based on a high-order filter according to claim 5, characterized in that: The infrared analysis model is a recurrent neural network model.

9. An electronic device, characterized in that: include: processor; Memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the motor control method based on the high-order filter according to any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the motor control method based on a high-order filter according to any one of claims 1 to 4 is implemented.

Images