在不同長(zhǎng)度尺度上具有特定建構(gòu)的分級(jí)材料在自然界中隨處可見，如骨骼、木材等。在結(jié)構(gòu)中增添建構(gòu)可以增強(qiáng)材料的機(jī)械性能，是對(duì)原子級(jí)微觀結(jié)構(gòu)和宏觀級(jí)零件維度進(jìn)行設(shè)計(jì)的又一手段。因此，對(duì)分級(jí)建構(gòu)材料的研究，包括控制材料的疲勞耐受性、能量吸收、剛度和強(qiáng)度等，引起了人們濃厚的興趣。蜂窩狀結(jié)構(gòu)材料由于其極低的重量和優(yōu)異的機(jī)械性能，在汽車、鐵路、航空航天工業(yè)中均有著廣泛的應(yīng)用。Fig. 1 Data representation of MD simulations.近年來(lái)，人工智能的發(fā)展增強(qiáng)了建構(gòu)化設(shè)計(jì)的能力，在實(shí)現(xiàn)受生物啟發(fā)的分級(jí)復(fù)合材料、使用圖神經(jīng)網(wǎng)絡(luò)的半監(jiān)督方法以及自然語(yǔ)言輸入生成設(shè)計(jì)架構(gòu)化材料等方面均取得了成功。同時(shí)，機(jī)器學(xué)習(xí)模型也常常已被應(yīng)用于其他材料性能的研究，如預(yù)測(cè)斷裂、柔度和屈曲等多種力學(xué)性能。Fig. 2 LSTM model training. a An ensemble of 1445 MD simulations were used to train the convolutional LSTM network. b Predicted ML stresses align well with real MD stresses, with an r2 = 0.95 and (c) validation loss = 0.00058. d Predicted curves across a range of stress behaviors align well with MD, with the samples from Fig. 1b provided as example.來(lái)自麻省理工學(xué)院原子和分子力學(xué)實(shí)驗(yàn)室的Andrew J. Lew等，提出了一個(gè)建構(gòu)化蜂窩狀材料壓縮設(shè)計(jì)的完整工作流程。他們使用分子動(dòng)力學(xué)模擬確定了分級(jí)蜂窩狀晶格空間，使用機(jī)器學(xué)習(xí)和遺傳算法生成了目標(biāo)行為的候選結(jié)構(gòu)，并利用增材制造技術(shù)對(duì)頂級(jí)候選結(jié)構(gòu)進(jìn)行快速測(cè)試。Fig. 3 Inverse design procedure. The stress prediction ML model directly solves the forward design problem, where we input an arbitrary structure vector and rapidly receive its stress strain curve. Here, we solve the inverse design problem via genetic algorithm, which comprises an iterative two stage process of generation and evaluation, to obtain structures given a desired stress behavior as input.訓(xùn)練后的機(jī)器學(xué)習(xí)模型為解決正向設(shè)計(jì)問題提供了一個(gè)有效的工具：對(duì)于給定的蜂窩狀超結(jié)構(gòu)，能夠直接快速預(yù)測(cè)其壓縮行為，而無(wú)需建立、運(yùn)行和分析物理模擬過程。他們通過模擬和實(shí)驗(yàn)，驗(yàn)證了遺傳算法搜索的有效性，可高效解決逆向設(shè)計(jì)問題。Fig. 4 Inverse design of stiffness and ultimate stress.作者的報(bào)道展示了一個(gè)從設(shè)想的性能需求到實(shí)際的材料結(jié)構(gòu)的“端到端”壓縮設(shè)計(jì)過程。該過程可以推廣到多種材料性質(zhì)，并且無(wú)需知道基材的特征。這為未來(lái)使用計(jì)算模擬、人工智能和實(shí)驗(yàn)手段協(xié)同增強(qiáng)材料設(shè)計(jì)提供了另一種途徑。該文近期發(fā)布于npj Computational Materials 9: 80 (2023)。Fig. 5 Experimental verification of stiffness design.Editorial SummaryArchitected materials design: Simulation, Deep learning and ExperimentationHierarchical materials with specific architecture at different length scales are observed everywhere in nature, like in bone and wood. Adding architecture to structures can enhance mechanical properties and provides an extra design lever on top of atomic-level microstructure and macroscopic-level part dimensions. Investigations into hierarchically architected materials have thus been of great interest, with efforts to control fatigue tolerance, energy absorption, and stiffness and strength, among many others. Honeycomb structures are of particular interest due to their ultra-low weight and outstanding mechanical properties, with a variety of applications across automotive, railway, and aerospace industries. Recent advances in artificial intelligence have afforded emerging capabilities for architectural design. For example, there have been successes in achieving bioinspired hierarchical composites, in using semi-supervised approaches with graph neural networks, and in implementing natural language inputs for generative design of architected materials. Concurrently, machine learning (ML) models have been used in other material platforms for the prediction of a multitude of mechanical properties including fracture, compliance, and buckling. Fig. 6 Experimental verification of stress design.Andrew J. Lew et al. from the Laboratory for Atomistic and Molecular Mechanics (LAMM), Massachusetts Institute of Technology,demonstrated a full workflow to tackle compression design of architected honeycomb materials. They used molecular dynamics simulations to determine initial insights into the space of hierarchical honeycomb lattices, machine learning and genetic algorithms to generate candidates for desired behavior, and additive manufacturing to rapidly test top structural candidates. The trained ML model provides an effective tool for the forward design problem, in which a given super-honeycomb structure can have its compressive behavior directly and rapidly predicted without having to set up, run, and analyze a physics-based simulation. A genetic algorithm search validated by simulation and experimentation enables effective interrogation of the inverse design problem. This work demonstrates a successful end-to-end process for compression design from ideated property requirements to actualized material structures. This process is generalizable to multiple material properties and agnostic to the identity of the base material, which can provide alternative avenues at the intersection of simulation, artificial intelligence, and experiment that can synergistically empower materials design in the future. This article was recently published in npj Computational Materials 9: 80 (2023).原文Abstract及其翻譯Designing architected materials for mechanical compression via simulation, deep learning, and experimentation (機(jī)械壓縮建構(gòu)化材料設(shè)計(jì)：計(jì)算模擬、深度學(xué)習(xí)和實(shí)驗(yàn)驗(yàn)證)Andrew J. Lew,Kai Jin & Markus J. Buehler Abstract Architected materials can achieve enhanced properties compared to their plain counterparts. Specific architecting serves as a powerful design lever to achieve targeted behavior without changing the base material. Thus, the connection between architected structure and resultant properties remains an open field of great interest to many fields, from aerospace to civil to automotive applications. Here, we focus on properties related to mechanical compression, and design hierarchical honeycomb structures to meet specific values of stiffness and compressive stress. To do so, we employ a combination of techniques in a singular workflow, starting with molecular dynamics simulation of the forward design problem, augmenting with data-driven artificial intelligence models to address the inverse design problem, and verifying the behavior of de novo structures with experimentation of additively manufactured samples. We thereby demonstrate an approach for architected design that is generalizable to multiple material properties and agnostic to the identity of the base material.與普通材料相比，建構(gòu)化材料能夠?qū)崿F(xiàn)增強(qiáng)的性能。材料中特定的建構(gòu)可以作為一種有力的設(shè)計(jì)手段，在不改變基材的情況下實(shí)現(xiàn)目標(biāo)行為。因此，材料建構(gòu)和對(duì)應(yīng)性能之間的聯(lián)系仍然是航空航天、民用工業(yè)、汽車應(yīng)用等眾多領(lǐng)域所感興趣的開放話題。這里，我們關(guān)注與機(jī)械壓縮相關(guān)的性能，設(shè)計(jì)了一種分級(jí)蜂窩狀結(jié)構(gòu)，以滿足特定剛度和壓縮應(yīng)力的需要。為此，我們?cè)趩蝹€(gè)工作流程中采用組合策略：從分子動(dòng)力學(xué)模擬出發(fā)解決正向設(shè)計(jì)問題，增加數(shù)據(jù)驅(qū)動(dòng)人工智能模型以解決逆向設(shè)計(jì)問題，并通過實(shí)驗(yàn)上制造的樣品驗(yàn)證了新結(jié)構(gòu)的機(jī)械行為。由此，我們給出了一種建構(gòu)化設(shè)計(jì)方法，該方法可推廣到多種材料性質(zhì)，并且無(wú)需知道基材的特征。【做計(jì)算 找華算】華算科技專注DFT代算服務(wù)、正版商業(yè)軟件版權(quán)、全職海歸計(jì)算團(tuán)隊(duì)，10000+成功案例！Nature Catalysis、JACS、Angew.、AM、AEM、AFM等狂發(fā)頂刊，好評(píng)如潮！計(jì)算內(nèi)容涉及OER、HER、ORR、CO2RR、NRR自由能臺(tái)階圖、火山理論、d帶中心、反應(yīng)路徑、摻雜、缺陷、表面能、吸附能等。添加下方微信好友，立即咨詢：
-透明背底.png)
