Method and system for optimizing the design of a structural component in a manufacturing environment

Abstract

Described herein is a method and optimization system (201) for performing design optimization of a component in a manufacturing environment (200). The method (600) includes receiving an initial design of a component to be manufactured along with corresponding input parameters. The method further includes applying an optimization technique (403) to the initial design based on the input parameters to produce an optimized design (405) of the component. The method further includes generating heat maps (311) corresponding to the input parameters using finite analysis. Thereafter, the method includes providing the heat maps (311) and the optimized design of the component to a neural network model (409).The neural network model generates a plurality of intermediate designs of the components in an iterative manner using the heat maps (311) until a final intermediate design matches the optimized design (405) of the component, and learns about the generation of the optimized design of the component.

Description

PREAMBLE FOR DESCRIPTION:In the following description, the invention and the manner in which it is practiced will be described in more detail.TECHNICAL FIELDThe present subject matter relates generally to methods of constructing a structural member, and more particularly, but not exclusively, to a method and system for performing structural optimization of a structural member in a manufacturing environment.BACKGROUND OF THE DISCLOSUREIn general, in the automobile industry, it is necessary to construct structural members according to requirements of a vehicle. Structural component construction includes several steps, such as analyzing the load on components and determining the shape and size of the components to allow the components to withstand the load. The design process for structural components is therefore a long-lasting iterative process, since it comprises various operations, such as creating a design, performing a simulation of the design, determining errors, correcting the determined errors, and repeating the simulation process until the desired shape of the design is reached. Once the design is present, the design of the structural members must be optimized to achieve better manufacturing results. Currently, optimization methods are generally used for optimizing the construction of structural components, such as, for example, bidirectional evolutionary structural optimization (BESO) and solid isotropic material with Penalty (SIMP). The BESO method is a finite element optimization method. In the BESO structure optimization method, an inefficient material is iteratively removed from a structure and at the same time an efficient material is added to the structure. The BESO method provides robustness and mullability of the final topology. In the SIMP optimization method, the material distribution is predicted for a design space, load conditions, constraints, and manufacturing constraints. In the SIMP structure optimization method, as shown in Fig. 1, first, a density is uniformly distributed as indicated in Step 1. The uniform density distribution is equal to a certain volume fraction. In step 2, the optimization loop begins with the arrangement and solution of the equilibrium equations using finite element analysis (FEA). In step 3, the derivatives of the perception function with respect to the design variables are estimated by means of sensitivity analysis. In step 4, an optional filtering technique is applied to solve a problem in design. In step 5, the design variables are updated based on either optimality criteria or moving asymptotics. In step 6, the updated design variables and the resultant topology are analyzed, and the analysis and optimization process is repeated until convergence is achieved, i.e., the analysis and optimization is repeatedly performed until the desired shape or final topology of the design is achieved, as indicated in step 7. The SIMP optimization method provides robustness. Therefore, SIMP can be used for any combination of design constraints. In the SIMP optimization method, a penalty can be set freely, and accordingly, an optimum penalty can be used. Although the BESO optimization method and the SIMP optimization method offer certain advantages, the BESO and SIMP methods also have some disadvantages. For example, the BESO and SIMP optimization techniques require high computational power to optimize design. Likewise, the BESO and SIMP optimization methods require more time to optimize the construction of the structural components.Some of the existing non-patent literature relates to structural optimization of the design. For example, in the non-patent literature "3D design using generative adventrial networks and physics-based validation", a self-updating generative construction model with a physics simulation is used, and a generative adventrial network (GAN) is used for the structure design in the form of point clouds. The said non-patent literature contains more noise, as it uses traditional or conventional generative networks. Another non-patent literature entitled "Topology Optimization Methods for 3D Structural Problems: a comparative study" discusses the use of various optimization methods which are based entirely on physical simulations and do not contain neural networks. Consequently, the optimization method in the non-patent literature mentioned is time-consuming and computationally intensive. Another non-patent literature, "network, system and method for 3D shape generation", discusses the use of an image network, the Visual Geometry Group Net (VGGN). This non-patent literature contains no details of structural optimization. The other non-patent literature, "Learning hybrid (surface-based and volume-based) shape representation", discloses a method for generating 3D representations from 2D images. This non-patent literature uses a neural network trained to predict a 3D surface that matches the actual 3D shape of the object, which is calculated using an implicit function. Finally, the non-patent literature "HandVNet: Deep voxel-based network for 3D hand shape and pose estimation from a single depth map" (depth voxel-based network for 3D hand shape and pose estimation from a single depth map) discloses an architecture for creating a 3D hand shape from the voxelized depth maps. These depth maps are converted to "joint" heat maps and then used in various neural networks to ultimately produce the shape of a hand. Said non-patent literature is very specific for creating a 3D hand shape from the depth maps and cannot be used for other applications. In view of the above, there is a need to optimize the construction of a component in a manufacturing environment.The information disclosed in this Background of the Disclosure section is only for enhancement of understanding of the general background of the invention and should not be taken as an acknowledgement or an indication that this information forms the prior art already known to a person skilled in the art.SUMMARY OF THE DISCLOSUREDescribed herein is a method of performing structural optimization of a component in a manufacturing environment. The method includes receiving an initial construction of a component to be manufactured along with corresponding input parameters including constraints and load conditions applicable to the component. Further, the method includes applying an optimization technique to the initial design based on the input parameters to produce an optimized design of the component. In addition, the method includes generating heat maps corresponding to the input parameters using finite element analysis. The method then includes providing the heat maps and optimized construction of the component to a neural network model. The neural network model may generate a plurality of intermediate designs of the components in an iterative manner using the thermal maps until a final intermediate design matches the optimized design of the component, and learn about the generation of the optimized design of the component based on the generation of the plurality of intermediate designs in the iterative manner.Further, the present disclosure relates to an optimization system for performing design optimization of a component in a manufacturing environment. The control system includes a processor and a memory. The processor is configured to receive an initial construction of a component to be manufactured along with corresponding input parameters including constraints and load conditions applicable to the component. Further, the processor is configured to apply an optimization method to the initial construction based on the input parameters to generate an optimized construction of the component. Moreover, the processor is configured to generate thermal maps corresponding to the input parameters using finite element analysis. Thereafter, the processor is configured to route the thermal maps and optimized construction of the device to a neural network model. The neural network model may generate a plurality of intermediate designs of the components in an iterative manner using the thermal maps until a final intermediate design matches the optimized design of the component, and may also learn about the generation of the optimized design of the component based on the generation of the plurality of intermediate designs in the iterative manner.The foregoing summary is illustrative only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, other aspects, embodiments and features will become apparent by reference to the drawings and the following detailed description.BRIEF DESCRIPTION OF THE DRAWINGSThe accompanying drawings, which form a part of this disclosure, illustrate exemplary embodiments and together with the description, explain the principles disclosed. In the figures, the leftmost digit(s) of a reference numeral indicates the figure in which the reference numeral first appears. In the figures, the same numbers are used to refer to like features and components. Some embodiments of systems and / or methods according to embodiments of the present subject matter will now be described, by way of example only, and with reference to the accompanying figures, in which: FIG. 1 is a flow diagram illustrating a method 100 for applying an SIMP optimization technique to optimize the construction of a component; FIG. 2 illustrates a manufacturing environment 200 for optimizing the construction of a component, in accordance with some embodiments of the present disclosure; FIG. 3 ashows a detailed block diagram 300 aof an optimization system 201 illustrated in FIG. 2, according to some embodiments of the present disclosure; FIG. 3 b shows a diagram 300 bof a heat map(s) 311 created based on the input parameters in accordance with some embodiments of the present disclosure; FIG. 4 ashows a detailed diagram 400 aof a proposed architecture for performing the design optimization of a component in a manufacturing environment 200, in accordance with some embodiments of the present disclosure. FIG. 4 b shows a detailed diagram 400 bfor the training of a neural network architecture 409 for generating intermediate constructs 411, in accordance with some embodiments of the present disclosure. FIG. 4 c shows a detailed diagram 400 cof the U-network convolutional architecture 409 with an encoder 413 and a decoder 415 for generating an intermediate construction 411 in accordance with some embodiments of the present disclosure. FIG. 5 shows a diagram 500 for estimating a loss function for each iteration of the production of an intermediate structure 411 of the structural component until the final intermediate structure matches the produced optimized structure 405, in accordance with some embodiments of the present disclosure. FIG. 6 is a flow diagram illustrating a method 600 for optimizing the construction of a component in a manufacturing environment 200, in accordance with some embodiments of the present disclosure.Those skilled in the art should appreciate that all block diagrams included herein represent conceptual views of systems embodying the principles of the present subject matter. It will also be understood that all flowcharts, flowcharts, state transition diagrams, pseudo code, and the like represent various processes that are substantially embodied in a computer readable medium and that can be executed by a computer or processor, regardless of whether such a computer or processor is explicitly illustrated.DETAILED DESCRIPTIONAs used herein, the word "exemplary" is used in the sense of "serving as an example, unit, or illustration.". Any embodiment or implementation of the present subject matter described herein as "exemplary" is not necessarily to be considered preferred or advantageous over other embodiments. Although the disclosure is susceptible to various modifications and alternative forms, a specific embodiment thereof has been shown by way of example in the drawings and will be described in detail below. However, it is not intended to limit the disclosure to the specific forms, but on the contrary, the disclosure is intended to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure. The terms "comprise," "comprise," "includes," or other variations thereof, are intended to cover a non-exclusive inclusion, such that a structure, apparatus, or method comprising a list of components or steps not only includes those components or steps, but may also include other components or steps not expressly listed or associated with such structure or apparatus or method. In other words, one or more elements in a system or device initiated with "comprises... a" does not exclude, without further limitations, the presence of other elements or additional elements in the system or method. In the following detailed description of the embodiments of the disclosure, reference is made to the accompanying drawings, which form a part hereof, and in which is shown by way of illustration specific embodiments in which the disclosure may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the disclosure, and it is to be understood that other embodiments may be utilized and that changes may be made without departing from the scope of the present disclosure. The following description is therefore not to be taken in a limiting sense.FIG. 2 illustrates a manufacturing environment 200 for performing optimization of the construction of a component, in accordance with some embodiments of the present disclosure. As shown in FIG. 2, manufacturing environment 200 may include an optimization system 201 and an internal unit 209. The manufacturing environment 200 may include, for example, an automobile industry, a first-out truck, or an industry for manufacturing components. The manufacturing environment 200 may be other environments in which the structural members are constructed and manufactured. The optimization system 201 may be any computer system capable of performing structural optimization of a structural component in the manufacturing environment 200. The optimization system 201 may include an interface 203, a memory 205, and a processor 207. The optimization system 201 may be, for example, a desktop computer, a laptop computer, or any computing machine that may be used to construct the structural component. The internal entity 209 may be an internal construction team or an internal specification provider in the manufacturing environment. In another embodiment, the internal unit 209 may be a third party that provides design specifications for the manufacture of the structural members. The optimization system 201 may communicate with the internal unit 209 to obtain structural component design specifications for structural component design.Initially, the optimization system 201 may receive a first construction of a structural component to be manufactured from the internal unit 209. The initial construction of the structural member may be received in the form of a three-dimensional (3D) structure or a three-dimensional (3D) block. The optimization system 201 can also receive input parameters together with the initial construction of the structural component. In one example, a user may build the initial construction of the structural members and define the required input parameters. The input parameters can include, among other things, the load conditions, boundary conditions and secondary conditions applicable to the structural component. For example, if a vehicle door (i.e., structural member) is constructed, the load conditions for the vehicle door may be received. Similarly, the vehicle door constraints and constraints may be received. In one example, the constraints may correspond to the loads acting on the construction of the structural member. The load conditions exerted on the structural component can be, for example, 50 Newtons (N). The present disclosure is not limited thereto. Other values may also be used which correspond to the load conditions applicable to the construction of the structural member. The load conditions may be applied based on the construction of the structural member. For example, if the design of a vehicle suspension is involved, the load can be applied vertically. Also in other structural member constructions, the loads can be applied from different angles and at different values. In one example, the constraints may be the information relating to the structural component to be constructed. For example, the boundary conditions provide information about where the structural component will be fixed or freely movable during actual implementation and use.Once the initial construction of the component has been received, the optimization system 201 applies an optimization technique to the initial construction based on the input parameters to produce an optimized construction of the component. In one example, the optimized construction produced may be a real output of the structural component that may be produced using existing optimization techniques such as bidirectional evolutionary structure optimization (BESO), solid isotropic penalization (SIMP), and / or variational topology optimization (VARTOP) to produce the optimized construction. In one example, the optimization system 201 may apply the SIMP optimization technique to generate an optimized construction of the structural component. The present disclosure is not limited thereto. In one example, MATLAB-based SIMP topology optimization may be used to generate optimized designs or structures. Those skilled in the art will understand that the optimized design produced with the above technique requires various mechanical simulations of the original design. Performing such simulations requires additional computing power and time as described in the Background section of this specification. However, the present disclosure takes advantage of the existing technique and trains a neural network model. The neural network model can generate the optimized construction of the component after training without the need to use the existing optimization techniques, and therefore makes the construction process faster and more efficient. A more rapid and efficient construction of the components provides another advantage in the manufacturing cycle of the components. In the following sections of the specification, how the neural network model is trained within the scope of the present disclosure will now be explained in detail.After creating the optimized structural component construction using the existing optimization technique, the optimization system 201 may generate thermal maps according to the input parameters using finite element analysis (FEA). The FEA may use a finite element method (FEM) to generate the thermal maps. The FEM may be a numerical method which can solve edge value problems. For example, the FEM may subdivide a block into smaller elements or smaller parts referred to as finite elements. The finite elements are then analyzed to generate the thermal maps. The heat maps are shown in Figure 3b. In one example, the same input parameters as loads, boundary conditions and restrictions are used for the creation of the heat maps, which input parameters are also used in the creation of the optimized construction. The thermal maps may be an image or a representation of the image in the form of a map. The thermal maps may be individual values that can be represented in different colors in the thermal map. In one example, the thermal maps may provide information regarding displacement of the component in multiple axes, a stress applied to the component, an elongation applied to the component, and a volume fraction of the component. The thermal maps may include one or more channels representing node displacement in X, Y, and Z directions. Similarly, four channels of the one or more channels may represent the node strains εx, εy, εzand the shear strain xyz. Optionally, the thermal maps are pre-processed before being supplied to the neural network model. The preprocessing steps may include performing normalization of the heat maps to remove anomalies and adjusting size and shape to obtain heat maps of the same size and shape. In the next step, the optimization system 201 provides the component's thermal maps to the neural network model. The neural network model may contain random weight values when it receives the thermal maps. The neural network may generate a plurality of intermediate designs of the component in an iterative manner using the thermal maps until a final intermediate design matches the generated optimized design (or real output) of the component. In one example, the neural network model may be a U-network convolutional network model. The U-mesh neural network model may include an encoding layer and a decoding layer. The coding layer may include an encoder that captures or receives thermal maps of the device and then performs down-sampling through general convolution operations and then reduces the spatial resolution of the thermal maps. Alternatively, the decoding layer may obtain the thermal maps with reduced spatial resolution and perform up-sampling of the thermal maps by performing general convolution operations and concatenation operations and generating the plurality of intermediate constructs.During the iterative generation of the multiple intermediate designs of the component, the optimization system 201 compares the individual intermediate designs with the generated optimized design. Initially, the intermediate construction produced may be scrap, as the neural network is not trained. As a result, the intermediate construction generated may have significant differences compared to the optimized construction. Based on the comparison, the optimization system 201 calculates a loss function that indicates a design difference between the intermediate design and the generated optimized design. The loss function may be calculated using a sum of Kullback-Leibler (KL) divergence and an L2 regularization technique. The KL divergence may be a divergence technique used to calculate a loss or difference between a function and a reference function. L2 regularization may be a regularization technique that removes a small percentage of the weights on each iteration of the generation of intermediate constructs. The loss function can be calculated using the following formulas:KL Divergence Loss:P(x) is the optimized construction and Q(x) is the intermediate constructionL2 Regularization:Here, "θ i" is the parameter of the generator network, such as the convolution kernel.After the loss function is calculated at each comparison unit, the optimization system 201 performs backpropagation. During backpropagation, the calculated loss function is passed to the neural network model so that the intermediate construction produced approximates the optimized construction of the component at each iteration. To perform the backpropagation, the optimization system 201 may calculate gradients for the loss in net weight, and then the net weights may be updated or adjusted. The goal of such a process is to minimize L2loss (as explained in FIG. 5) by iteratively adjusting the weight parameters. In this way, the optimization system 201 helps the neural network model learn to create the optimal construction of the component based on iteratively generating a plurality of intermediate designs. Then, the optimization system 201 may apply the learning to the obtained initial construction for a component to be subsequently manufactured to generate the corresponding optimized construction. This makes it possible to avoid the use of optimization methods for producing the corresponding optimized construction. If the intermediate construction produced matches the optimized construction produced, the process may be terminated. Those skilled in the art will understand that the optimized construction of the component includes a number of dimensional values of the component that allow the component to withstand the load conditions. Once the optimized design is established and completed, the original design can be altered by adding and / or removing materials from the component such that the optimized design of the components can withstand the loading conditions.FIG. 3 shows a detailed block diagram 300 of an optimization system 201 shown in FIG. 2, in accordance with some embodiments of the present disclosure. In one embodiment, optimization system 201 may include an interface 203, memory 205, and processor 207 (also referred to as "CPUs" or "the one or more processors"). In some embodiments, the memory 205 may be communicatively coupled to the one or more processors 207. The memory 205 stores instructions executable by the one or more processors 207. The one or more processors 207 may include at least one data processor for executing components for executing user- or system-generated requests. The one or more processors 207 may perform one or more functions of the optimization system 201 to perform structural optimization of a component in a manufacturing environment. The memory 205 may store instructions executable by the one or more processors 207 that, when executed, may cause the one or more processors 207 to perform component design optimization in a manufacturing environment. The memory 205 may also store the initial designs and corresponding input parameters of various components to be designed and manufactured. The interface 203 may be coupled to the one or more processors 207 via which the required specifications for the design are received. For example, the one or more processors 207 may communicate with an internal unit 209 to receive the design specifications, as shown in FIG. 2. In one embodiment, the one or more processors 207 may include one or more modules or hardware units, e.g., a receiving unit 301, an application unit 303, a generating unit 305, a providing unit 307, and a learning unit 309, but not just these. In some embodiments, the one or more modules or units may be software modules that may be stored in the memory 205. The one or more modules or hardware units may be configured to perform the various operations of the present disclosure to perform structural optimization of a component in a manufacturing environment.FIG. 3 b shows a diagram 300 bof a heat map 311 created based on the input parameters in accordance with some embodiments of the present disclosure. According to the present disclosure, the heat map 311 is generated based on the same input parameters as loads, constraints, and constraints used in the creation of the optimized design. As already explained, the thermal maps 311 may be an image or a representation of the image in the form of a map. The thermal maps 311 provide information about the displacement 313 of the component in different axes, a stress 315 applied to the component, an elongation applied to the component, and a volume fraction of the component.FIG. 4 ashows a detailed diagram 400 aof the proposed architecture for performing the design optimization of a component in a manufacturing environment 200, according to some embodiments of the present disclosure. Initially, initial construction of a component 401 and input parameters may be received from the internal unit 209. The input parameters include design specification parameters (e.g., loads, constraints, and constraints) for the design of the structural member. After receiving the initial component 401 and the input parameters, the existing optimization methods 403, such as BESO, SIMP and / or VARTOP, may be used to generate the first output 405. The first output 405 may be, for example, a true optimized design generated with the existing optimization methods 403. Similarly, the same input parameters used in creating the real optimized construction may also be used for creating thermal maps 311. The FEM method can be used to create the heat maps. After the creation of the thermal maps 311, the thermal maps 311 may be provided to the neural network architecture or neural network model 409. The neural network architecture 409 may be a U-network convolutional network architecture. When the thermal maps 311 are provided, the architecture of the neural network 409 may generate a second output 411. The second output 411 may be an intermediate construction of the structural component. The second output 411 that is initially generated may be scrap and does not match the actual optimized design or first output 405 because the neural network architecture 409 is not trained and the neural network architecture 409 includes random weights. Subsequently, the second output 411 may be compared to the first output 405. In other words, the actual optimized construction 405 (first output) and the intermediate construction (second output) may be compared with a discriminator 407. In one example, the discriminator 407 may be a comparator used to compare the intermediate construction 411 and the real optimized construction 405 of the structural component. The discriminator 407 may be trained to compare the generated intermediate construct 411 to the optimized construct 405 and estimate the loss based on the comparison results. Additionally, the discriminator 407 may estimate a loss function by comparing the real optimized construct 405 and the intermediate construct 411. The calculated loss function may be provided to the neural network architecture 409 for training the neural network. During each iteration, the architecture of the neural network 409 may learn to generate the intermediate construction 411 with fewer losses than the previous iteration of generating the intermediate construction. The number of iterations may be performed until the output obtained (i.e., the intermediate construction) matches the optimized construction 405 generated. If the output obtained matches the generated optimized construct 405, the process may be terminated.FIG. 4 b shows a detailed diagram 400 bfor training a neural network architecture 409 to generate an intermediate construction 411 in accordance with some embodiments of the present disclosure. As shown in FIG. 4 b, thermal maps 311 may be provided as input to neural network architecture 409. Neural network architecture 409 is a U-neural network. The input to neural network architecture 409 may have a spatial resolution of 64*64*64 and 8 channels. The present disclosure is not limited thereto. The neural network may receive the thermal maps 311 and perform normal convolution functions, such as down sampling and up sampling images, and then generate an output in an output layer 411. The output of neural network architecture 409 may be a 3D construction of the structural component. The loss can be calculated by comparing the output construct to the optimized construct 405 and estimating the loss. The output dimensions of the 3D construction may be the same as the input dimensions (64*64*64). The output can be in voxel format. In FIG. 4 c, the functionality of the architecture of the neural network 409 is explained in detail.FIG. 4 c shows a detailed diagram 400 cof the convolutional U-network neural architecture 409 with an encoder 413 and a decoder 415 for generating an intermediate construction 411 in accordance with some embodiments of the present disclosure. As shown in FIG. 4 c, the U-neural network architecture 409 may include the encoder 413, the decoder 415, and a bottleneck layer 417. The encoder 413 can process 8 channels as an input. Each channel may represent a different thermal map 311 with a voxel resolution of 64*64*64. The encoder 413 can process the received input by a combination of convolutional layers and max pooling operations and then reduce the spatial dimensions of the input. Thereafter, the encoder 413 may generate a feature map having 512 channels and a spatial resolution of 8*8*8. This reduced feature map may contain important relationships and features of the original input. The feature map with the 512 channels and the spatial resolution of 8*8*8 may be stored in the Bottleneck layer 417. The Bottleneck layer 417 may be a bridge between the encoder 413 and the decoder 415. The Bottleneck layer 417 may allow the neural network architecture 409 to detect relevant features and discard less important details of the input. The decoder 415 may receive the feature map of the Bottleneck layer and performs up-sampling of the feature map to obtain the original input resolution of 64*64*64. Decoder 415 may achieve this by performing a series of up-sampling layers, convolution operations, and concatenation. The upsampling layers may increase the spatial dimensions and the convolutional layers may process the features and gradually restore the important spatial details or information. At the end of decoder 415, output layer 411, which is a final convolutional layer, may be used. The last convolutional layer can reduce the number of channels to 1 and generate a single channel 3D structure. The output of this layer may be a binary mask in which each voxel indicates the presence or absence of a target structure or a generated optimized construction. The 3D structures of the output layer 411 may be represented as voxel grids. Each voxel in the grid may be a binary value that indicates the presence (i.e., 1) or absence (i.e., 0) of a structure at that point in 3D space.FIG. 5 shows a diagram 500 for estimating a loss function for each iteration ( 501 a, 501 b,..., 501 n) of generating intermediate structures 411 of structural components until the final intermediate structure matches the generated optimized structure 405, in accordance with some embodiments of the present disclosure. In FIG. 5, the intermediate structures are represented by the reference numerals 502 a, 502 b...502 nfor each iteration. Similarly, optimal designs are represented by reference numerals 503 a, 503 b...503 nfor each iteration, and losses are represented by reference numerals 504 a, 504 b...504 nfor each iteration. As shown in FIG. 5, prior to the beginning of an iteration 1 or the first iteration 501 a, an optimized construction 405 of the structural component can be created on the basis of input parameters using optimization methods 403 existing in BESO or SIMP. During iteration 1 501 a, heat maps 311 are created based on the same input parameters and supplied to neural network architecture 409. The neural network architecture 409 may generate the intermediate construction 1 502 a. The intermediate construction 1 502 acan be a scrap construction. The intermediate construction 1 502a may be compared to the optimized construction 1 503a using the discriminator 407 (see Figure 4a). Discriminator 407 may compare voxel-to-voxel features of both intermediate construction 1 and optimized construction 405 and estimate loss 504a for the first iteration. Specifically, the discriminator 407 may calculate the L2 loss and update the weights in the output layer 411 using the L2 loss. The formula for the estimation of the L2loss may be: (1 / N)*Σ(intermediate design-optimized design). 2The steps for calculating the L2loss are as follows:Here, "η" is a learning rate. The value of "η" may be 0.1. The present disclosure is not limited thereto.Here, ∂L2 / ∂W is a gradient of the L2loss in the output layer 411.The estimated or calculated loss may be fed into the neural network architecture 409 by backpropagation. In one example, the L2loss in the first iteration 501 amay be 0.98. The present disclosure is not limited thereto. At the next iteration, the weights in neural network architecture 409 may be updated for each individual block. The L2loss may be continuously estimated and provided to the architecture of the neural network 409. In one example, the L2loss estimated in the second iteration may be less than the L2loss estimated in the first iteration. This means that the neural network architecture 409 learns to generate an intermediate construction that approximates the actual optimized construction 405. Similarly, in an iteration 10 or tenth iteration, intermediate construction 10 may be compared to optimized construction 405 using discriminator 407 (shown in FIG. 4 a). Discriminator 407 may compare voxel-to-voxel characteristics of both intermediate construct 10 and optimized construct 405 and estimate loss for the tenth iteration. The L2loss may be estimated for the tenth iteration. In one example, the L2loss in the tenth iteration may be 0.63. The present disclosure is not limited thereto. Finally, in an iteration 'n' or n th iteration, the intermediate construction 'n' may be compared to the optimized construction 405 using the discriminator 407 (as shown in Figure 4a). Discriminator 407 may compare voxel-to-voxel features of both intermediate construction 'n' and optimized construction 405 and estimate a loss for iteration n th. The L2loss may be estimated for n th iterations. In one example, the L2loss in the n th iteration may decrease to 0.02. As can be seen, the L2loss is less compared to the previous iterations. This indicates that the intermediate construction produced is a final intermediate construction and matches the real optimized construction 405 produced. In an example, the n th iteration may be the iteration 100. The present disclosure is not limited thereto. The L2loss may be reduced even before the 100 th iteration, and then the final intermediate construction may be obtained before the 100 th iteration.FIG. 6 shows a flow diagram illustrating a method 600 for performing structural optimization of a component in a manufacturing environment, in accordance with some embodiments of the present disclosure. As shown in FIG. 6, the method 600 may include one or more steps. The method 600 may be described in the general context of computer-executable instructions. In general, computer-executable instructions may include routines, programs, objects, components, data structures, procedures, modules, and functions that perform particular functions or implement particular abstract data types. The order in which method 600 is described is not intended to be limiting, and any number of the described method blocks may be combined in any order to implement the method. Moreover, individual blocks may be deleted from the methods without compromising the scope of the subject matter described herein. Moreover, the method may be implemented in any suitable hardware, software, firmware, or combination thereof.At block 602, the method 600 includes receiving an initial construction of a component to be manufactured along with corresponding input parameters including constraints and load conditions applicable to the component. The operations of the block 602 may be performed by the processor 207 (in particular, by the receiving unit 301) of FIG. 3. At block 604, the method includes applying an optimization method to the original design based on the input parameters to generate an optimized design of the component. The operations of block 604 may be performed by the processor 207 (specifically, by the application unit 303) of FIG. 3. At block 606, the method 600 includes generating heat maps corresponding to the input parameters using finite element analysis. The operations of block 606 may be performed by processor 207 (in particular, by generation unit 305) of FIG. 3. At block 608, the method 600 includes providing the thermal maps and optimized construction of the component to a neural network model 409. The operations of block 608 may be performed by processor 207 (in particular, by provisioning unit 307) of FIG. 3. At block 610, the method 600 includes generating a plurality of intermediate designs of the component in an iterative manner using the thermal maps until a final intermediate design matches the optimized design of the component. The operations of the block 610 may be performed by the processor 207 (in particular, by the generation unit 305) of FIG. 3. At block 612, the method 600 includes learning to generate the optimized construction of the component based on generating the plurality of intermediate constructions in the iterative manner. The operations of block 612 may be performed by processor 207 (specifically, learning unit 309) of FIG. 3.The advantages of the embodiments of the present disclosure are presented herein.In one embodiment, the present disclosure uses a neural network model that learns and generates an optimized construction in real-time. When the user or provider of specifications provides the design specifications, the present disclosure may generate the heat maps and provide the heat maps to the neural network model. Since the neural network is already trained for the various thermal maps, the present disclosure provides the required optimized design in a very short time. For example, the present disclosure may provide the required optimized design in a few minutes as compared to existing structure optimization techniques that require more time to build the optimized design. This greatly reduces the time required to create the optimized construction compared to the existing methods. Consequently, the required computing power can be efficiently reduced.In one embodiment, the present disclosure creates multiple optimized structures of the structural members. As a result, the present disclosure uses or utilizes minimal computational resources (e.g., minimal battery power) for generating optimized designs. Thus, the present disclosure requires less computing power and efficiently increases the performance of the system. In one embodiment, the present disclosure may be used in place of existing structure optimization methods such as BESO, SIMP, etc.The terms "an embodiment," "an embodiment," "embodiments," "the embodiment," "the embodiments," "one or more embodiments," "some embodiments," and "an embodiment" mean "one or more (but not all) embodiments of the invention(s)," unless expressly stated otherwise. The terms "including", "comprising", "with", and variations thereof mean "including, but not limited to" unless expressly stated otherwise. The enumeration of items does not mean that all or individual items are mutually exclusive unless expressly stated otherwise. The terms "a", "an" and "the" mean "one or more" unless expressly stated otherwise. The description of an embodiment having a plurality of interconnected components does not mean that all of these components are required. Rather, a variety of optional components are described to illustrate the wide variety of possible embodiments of the invention. When a single device or article is described herein, it will be appreciated that more than one device / article (whether cooperating) may be used in place of a single device / article. When more than one device or article is described herein (whether cooperating), it is understood that a single device or article may be used in place of more than one device or article, or that a different number of devices or articles may be used in place of the specified number of devices or programs. The functionality and / or features of a device may alternatively also be embodied by one or more other devices not expressly described as being equipped with that functionality / features. Therefore, other embodiments of the invention need not include the device itself. Finally, the language used in the specification has been chosen primarily for readability and guidance purposes, and not to delineate or rewrite the subject matter. It is therefore intended that the scope of the invention be limited not by this detailed description, but rather by all claims that rely on an application based thereon. Accordingly, the embodiments of the present invention are intended to be illustrative, but not limiting, of the scope of the invention as set forth in the following claims. While various aspects and embodiments have been disclosed herein, other aspects and embodiments will be apparent to those skilled in the art. The various aspects and embodiments disclosed herein are illustrative and not restrictive, the true scope and spirit of the invention being indicated by the following claims.

Claims

A method (600) of performing a design optimization of a component in a manufacturing environment (200), the method (600) comprising: receiving, by an optimization system (201), an initial design of a component to be manufactured along with corresponding input parameters comprising constraints and loading conditions applicable to the component; applying, by the optimization system (201), an optimization method (403) to the original design based on the input parameters to generate an optimized design (405) of the component; generating, by the optimization system (201), thermal maps (311) corresponding to the input parameters using finite element analysis; providing the thermal maps (311) and the optimized construction (405) of the component by the optimization system (201) to a neural network model (409), wherein the neural network model (409): generates a plurality of intermediate constructions of the component in an iterative manner using the thermal maps (311) until a final intermediate construction matches the optimized construction (405) of the component; and learns about the generation of the optimized construction (405) of the component based on the generation of the plurality of intermediate constructions in the iterative manner.The method (600) of claim 1, further comprising: applying the learning to a subsequent initial construction received for a component to be subsequently manufactured to produce a corresponding subsequent optimized construction, thereby preventing use of the optimization technique to produce the corresponding subsequent optimized construction; and modifying, based on the learning, the initial construction by performing at least one of adding and removing a material from the component such that the optimized construction of the component resists the loading conditions, wherein the initial construction of the component is received in the form of a three-dimensional block, and wherein the optimized construction (405) of the component comprises a set of dimensional values of the component that allow the component to resist the loading conditions.The method (600) of claim 1, wherein generating a plurality of intermediate designs of the component occurs in an iterative manner using the thermal maps (311) until a final intermediate design matches the optimized design (405) of the component: comparing each of the plurality of intermediate designs to the optimized design (405); calculating a loss function based on the comparison, the loss function indicating a design difference between the intermediate design and the optimized design (405); and performing backpropagation using the loss function on the neural network model (409) such that at each iteration, the generated intermediate design (411) comes closer to the optimized design (405) and the generated final intermediate design matches the optimized design (405) of the component.The method (600) of claim 3, wherein performing backpropagation using the loss function on the neural network model (409) further comprises: adjusting network weights of the neural network model (409) at each entity when the loss function is calculated and input to the neural network model (409), based on the comparison of the intermediate construction (411) and the optimized construction (405).The method (600) of claim 1, wherein the thermal maps (311) comprise information about a displacement (313) of the component in multiple axes, a stress (315) applied to the component, an elongation applied to the component, and a volume fraction of the component.An optimization system (201) for performing a structural optimization of a component in a manufacturing environment (200), the optimization system (201) comprising: a processor (207); and a memory (205) coupled to the processor (207), the processor (207) configured to: obtain an initial construction of a component to be manufactured along with the corresponding input parameters comprising constraints and loading conditions applicable to the component; apply an optimization method (403) to the original construction based on the input parameters to generate an optimized construction (405) of the component; generate thermal maps (311) corresponding to the input parameters using finite element analysis; providing the thermal maps (311) and the optimized construction (405) of the component for a neural network model (409), wherein the neural network model (409): generates a plurality of intermediate constructions of the component in an iterative manner using the thermal maps (311) until a final intermediate construction matches the optimized construction (405) of the component; and learns about the generation of the optimized construction (405) of the component based on the generation of the plurality of intermediate constructions in the iterative manner.The optimization system (201) of claim 6, wherein the processor (207) is further configured to: apply the learning to a subsequent initial construction received for a component to be subsequently manufactured to produce a corresponding subsequent optimized construction, thereby preventing use of the optimization technique (403) to produce the corresponding subsequent optimized construction; and modify the initial construction by performing at least one of adding and removing a material from the component, based on the learning, such that the optimized construction (405) of the component resists the loading conditions, wherein the initial construction of the component is received in the form of a three-dimensional block, and wherein the optimized construction (405) of the component comprises a set of dimensional values of the component that allow the component to resist the loading conditions.The optimization system (201) of claim 6, wherein the processor (207) is configured to generate a plurality of intermediate designs of the component in an iterative manner using the thermal maps (311) until a final intermediate design matches the optimized design (405) of the component: compare each of the plurality of intermediate designs with the optimized design (405); calculate a loss function based on the comparison, the loss function indicating a design difference between the intermediate design (411) and the optimized design (405); and perform backpropagation using the loss function on the neural network model (409) such that at each iteration, the generated intermediate design (411) approximates the optimized design (405) and the generated final intermediate design matches the optimized design (405) of the component.The optimization system (201) according to claim 8, wherein the processor (207) is configured to perform the backpropagation using the loss function on the neural network to: adjust the network weights of the neural network model (409) at each unit when the loss function is calculated and input to the neural network model (409), based on the comparison between the intermediate construction (411) and the optimized construction (405).The optimization system (201) of claim 6, wherein the thermal maps (311) comprise information about a displacement (313) of the component in multiple axes, a stress (315) applied to the component, an elongation applied to the component, and a volume fraction of the component.

Images