# US 2021/0241100 A1 · Apparatus and method for boundary learning optimization > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2021/0241100 A1 - Document ID: US20210241100A1 - Status: Published application - Kind: A1 - Application number: 17/139,800 - Filed: 2020-12-31 - Published: 2021-08-05 - Earliest claimed: 2018-07-21 - Assignees: The Regents of the University of California - Applicants: The Regents of the University of California - Inventors: Jonathan Hopkins; Ali Hatamizadeh; Yuanping Song - Web page: https://patentgrep.com/patent/US20210241100A1 ## CPC - G06N 3/08: Learning methods - G06N 3/04: Architecture, e.g. interconnection topology - G06N 3/0499: Feedforward networks - G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections - G06N 3/09: Supervised learning - G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A boundary learning optimization tool for training neural networks with accurate models of parameterized flexure using a minimal number of numerically generated performance solutions generated from different design instantiations of those topologies. Performance boundaries are output by the neural network in optimization steps, with geometric parameters varied from smallest allowable feature sizes to largest geometrically compatible feature sizes for given constituent materials. The plotted performance boundaries define the design spaces of flexure systems toward allowing designers to visually identify which geometric versions of their synthesized topologies best achieve desired combinations of performance capabilities. ## Claims ### Claim 1 An apparatus for boundary learning optimization on flexure systems, the apparatus comprising: (a) an artificial neural network having an input layer, multiple intermediate layers and an output layer; (b) a processor configured for training said artificial neural network and analyzing output from models of parameterized flexure from said artificial neural network; and (c) a non-transitory memory storing instructions executable by the processor; (d) wherein said instructions, when executed by the processor, perform one or more steps comprising: (i) training said artificial neural network to provide sufficiently accurate models of parameterized flexure using a minimal number of numerically generated performance solutions generated from different design instantiations of topologies; (ii) plotting a performance boundary of a desired topology using models output by said artificial neural network in an optimization step on said processor; and (iii) wherein said apparatus identifies the boundaries of performance capabilities achieved by general flexure system topologies if their geometric parameters are varied from smallest allowable feature sizes to largest geometrically compatible feature sizes for given constituent materials; (e) whereby boundaries of performance capabilities plotted by said apparatus fully define the design spaces of flexure systems toward allowing designers to visually identify which geometric versions of their synthesized topologies best achieve desired combinations of performance capabilities; and (f) wherein said apparatus trains neural networks for creating models based on the results of numerically generated data. ### Claim 2 (depends on 1) The apparatus of claim 1, wherein said apparatus is configured as a complementary tool for use with freedom and constraint topologies (FACT) synthesis approach, in which said apparatus optimizes geometry of flexure topologies synthesized using FACT. ### Claim 3 (depends on 1) The apparatus of claim 1, wherein said performance boundary of a desired topology circumscribes a combination of desired capabilities that can be achieved by every geometric instantiation of the desired topology without the need to calculate capabilities achieved by every design instantiation. ### Claim 4 (depends on 1) The apparatus of claim 1, wherein said plotting of a performance boundary comprises iterating using additional numerically or analytically generated solutions from each updated boundary that is generated until the final boundary is obtained to a given level of accuracy. ### Claim 5 (depends on 1) The apparatus of claim 1, wherein said numerically generated performance solutions comprise a training set and a testing set, and wherein said training set is configured for receipt by a neural network for training the neural network. ### Claim 6 (depends on 1) The apparatus of claim 1, wherein said neural network is configured with an initial-guess architecture. ### Claim 7 (depends on 1) The apparatus of claim 1, wherein said instructions when executed by the processor further perform one or more steps comprising training a single neural network for each desired output performance capability plotted. ### Claim 8 (depends on 1) The apparatus of claim 1, wherein said boundary learning optimization on flexure systems apparatus utilizes an automated method for training neural networks to rapidly create accurate models of general physical systems. ### Claim 9 (depends on 1) The apparatus of claim 1, wherein each of said performance capabilities is associated with a degree of freedom in the physical system. ### Claim 10 A method for boundary learning optimization on flexure systems, the method comprising: (a) training an artificial neural network to provide sufficiently accurate models of parameterized flexure using a minimal number of numerically generated performance solutions generated from different design instantiations of those topologies; (b) plotting a performance boundary of a desired topology using models output by said artificial neural network in an optimization step performed by a processor executing instructions stored in a non-transitory memory; and (c) identifying boundaries of performance capabilities achieved by general flexure system topologies if their geometric parameters are varied from smallest allowable feature sizes to largest geometrically compatible feature sizes for given constituent materials; (d) whereby performance boundaries plotted by said method fully define the design spaces of flexure systems toward allowing designers to visually identify which geometric versions of their synthesized topologies best achieve desired combinations of performance capabilities. ### Claim 11 (depends on 10) The method of claim 10, wherein said method is configured as a complementary tool for use with freedom and constraint topologies (FACT) synthesis approach, in which said apparatus optimizes geometry of flexure topologies synthesized using FACT. ### Claim 12 (depends on 10) The method of claim 10, wherein said performance boundary of a desired topology circumscribes a combination of desired capabilities that can be achieved by every geometric instantiation of the desired topology without calculating capabilities achieved by every design instantiation. ### Claim 13 (depends on 10) The method of claim 10, wherein said plotting of a performance boundary comprises iterating using additional numerically or analytically generated solutions from each updated boundary that is generated until the final boundary is obtained to a given level of accuracy. ### Claim 14 (depends on 10) The method of claim 10, wherein said numerically generated performance solutions comprise a training set and a testing set, and wherein said training set is configured for receipt by a neural network for training the neural network. ### Claim 15 (depends on 10) The method of claim 10, wherein said neural network is configured with an initial-guess architecture. ### Claim 16 (depends on 10) The method of claim 10, wherein said method performs training of a single neural network for each desired output performance capability plotted. ### Claim 17 (depends on 10) The method of claim 10, wherein said boundary learning optimization on flexure systems apparatus utilizes an automated method for training neural networks to rapidly create accurate models of general physical systems. ### Claim 18 (depends on 10) The method of claim 10, wherein each of said performance capabilities is associated with a degree of freedom in the physical system.