The design of alloys has typically involved adaptive experimental synthesis and characterization guided by machine learning models fitted to available data. Often a bottleneck for sequential design, e.g., by Bayesian Global Optimization (BGO), be it for self-driven or manual synthesis, is that the search space becomes intractable as the number of alloy elements and its compositions exceeds a threshold. Here we overcome this limitation by performing compositional design of alloys using reinforcement learning (RL) within a closed loop with manual synthesis and characterization. Moreover, the training efficiency is increased by incorporating uncertainty within the reward. Existing alloy data is often limited, however, with pre-training the agent can access regions of higher reward values more frequently. In addition, the experimental feedback enables the agent to gradually explore new regions with higher rewards, compositionally different from the initial dataset. We demonstrate this strategy by designing a phase change multicomponent alloy (Ti27.2Ni47Hf13.8Zr12) with the highest transformation enthalpy (ΔH) -37.1 J/g within the TiNi-based family of alloys from a space of over 2×108 candidates, although the initial training is only on a compact dataset of 112 alloys. The approach directly applies to processing conditions where the actions would be performed in a given order.
Abstract The problem that is considered is that of maximizing the energy storage density of Pb‐free BaTiO 3 ‐based dielectrics at low electric fields. It is demonstrated that how varying the size of the combinatorial search space influences the efficiency of material discovery by comparing the performance of two machine learning based approaches where different levels of physical insights are involved. It is started with physics intuition to provide guiding principles to find better performers lying in the crossover region in the composition–temperature phase diagram between the ferroelectric phase and relaxor ferroelectric phase. Such an approach is limiting for multidopant solid solutions and motivates the use of two data‐driven machine learning and design strategies with a feedback loop to experiments. Strategy I considers learning and property prediction on all the compounds, and strategy II learns to preselect compounds in the crossover region on which prediction is carried out. By performing only two active learning loops via strategy II, the compound (Ba 0.86 Ca 0.14 )(Ti 0.79 Zr 0.11 Hf 0.10 )O 3 is synthesized with the largest energy storage density ≈73 mJ cm −3 at a field of 20 kV cm −1 , and an insight into the relative performance of the strategies using varying levels of knowledge is provided.
An active learning strategy using sampling based on uncertainties shows the promise of accelerating the development of new materials. We study the efficiencies of the active learning iteration loop with different uncertainty estimators to find the “best” material in four different experimental datasets. We use a bootstrap approach aggregating with support vector regression as the base learner to obtain uncertainties associated with model predictions. If the bootstrap replicate number B is small, the variance estimated by the empirical standard error estimator is found to be close to the true variance, whereas the jackknife based estimators give an upward or downward biased estimation of variance. As B increases, the bias of the jackknife based estimators decreases and the variance estimated finally converges to the true one. Therefore, the empirical standard error estimator needs the least number of iteration loops to find the best material in the datasets, especially when the bootstrap replicate number B is small. Our work demonstrates that an appropriate Bootstrap replicate B is conducive to minimizing calculation costs during the materials property optimization by active learning.
The relationship between material properties and independent variables such as temperature, external field or time, is usually represented by a curve or surface in a multi-dimensional space. Determining such a curve or surface requires a series of experiments or calculations which are often time and cost consuming. A general strategy uses an appropriate utility function to sample the space to recommend the next optimal experiment or calculation within an active learning loop. However, knowing what the optimal sampling strategy to use to minimize the number of experiments is an outstanding problem. We compare a number of strategies based on directed exploration on several materials problems of varying complexity using a Kriging based model. These include one dimensional curves such as the fatigue life curve for 304L stainless steel and the Liquidus line of the Fe-C phase diagram, surfaces such as the Hartmann 3 function in 3D space and the fitted intermolecular potential for Ar-SH, and a four dimensional data set of experimental measurements for BaTiO3 based ceramics. We also consider the effects of experimental noise on the Hartmann 3 function. We find that directed exploration guided by maximum variance provides better performance overall, converging faster across several data sets. However, for certain problems, the trade-off methods incorporating exploitation can perform at least as well, if not better than maximum variance. Thus, we discuss how the choice of the utility function depends on the distribution of the data, the model performance and uncertainties, additive noise as well as the budget.
Abstract Accelerating the discovery of advanced materials is crucial for modern industries, aerospace, biomedicine, and energy. Nevertheless, only a small fraction of materials are currently under experimental investigation within the vast chemical space. Materials scientists are plagued by time-consuming and labor-intensive experiments due to lacking efficient material discovery strategies. Artificial intelligence (AI) has emerged as a promising instrument to bridge this gap. Although numerous AI toolkits or platforms for material science have been developed, they suffer from many shortcomings. These include primarily focusing on material property prediction and being unfriendly to material scientists lacking programming experience, especially performing poorly with limited data. Here, we developed MLMD, an AI platform for materials design. It is capable of effectively discovering novel materials with high-potential advanced properties end-to-end, utilizing model inference, surrogate optimization, and even working in situations of data scarcity based on active learning. Additionally, it integrates data analysis, descriptor refactoring, hyper-parameters auto-optimizing, and properties prediction. It also provides a web-based friendly interface without need programming and can be used anywhere, anytime. MLMD is dedicated to the integration of material experiment/computation and design, and accelerate the new material discovery with desired one or multiple properties. It demonstrates the strong power to direct experiments on various materials (perovskites, steel, high-entropy alloy, etc). MLMD will be an essential tool for materials scientists and facilitate the advancement of materials informatics.
It is predicted that the thickness of a coating has major effects on a substrate in terms of mechanical and thermal properties. In this study, an Al2O3-ZrO2-SiO2 slurry was prepared as a coating material, which formed an alumina-zirconia-mullite composite coating after sintering. The alumina-zirconia-mullite composite coating was coated on a zirconia substrate to generate compressive stress in the coating due to the mismatch of the coefficient of thermal expansion (CTE). A series of coated samples with different coating thicknesses from ∼10 μm to ∼200 μm were prepared to investigate the effects of coating thickness. The residual compressive stress, thermal conductivity, CTE, and Young's modulus of the coating material were determined by relative methods, and the flexural strength of the coated and uncoated samples was measured by 3-point bending. The strength of the coated samples was 1362.98 ± 30.29 MPa, which is a 54.07% enhancement compared to the uncoated samples. The thermal conductivity of the coated samples was also increased compared to that of the uncoated samples. For a given thickness of the substrate of 2 mm, there was an optimum thickness of the coating of 90 μm, which showed the greatest strength compared to the other samples. Coatings that were too thin or too thick did not show the best reinforcement. Moreover, the porosity of the coated samples was also determined and discussed in this study. Comparison samples without SiO2 were also manufactured, and their flexural strength and thermal conductivity were found to not be as good as the samples with SiO2.
Multiple Mobile Robots System (MMRS) has shown many attractive features in lots of real-world applications that motivate their rapid and wide diffusion.In MMRS, the Cooperative Localization (CL) is the basis and premise of its high-performance task.However, the statistical characteristics of the system noise should be already known in traditional CL algorithms, which is difficult to satisfy in actual MMRS because of the numerous of disturbances form the complex external environment.So the CL accuracy will be reduced.To solve this problem, an improved Adaptive Active Cooperative Localization (A 2 CL) algorithm based on information optimization strategy for MMRS is proposed in this manuscript.In this manuscript, an adaptive information fusion algorithm based on the variance component estimation under Extended Kalman filter (VCEKF) method for MMRS is introduced firstly to enhance the robustness and accuracy of information fusion by estimating the covariance matrix of the system noise or observation noise in real time.Besides, to decrease the effect of observation uncertainty on CL accuracy further, an observation optimization strategy based on information theory, the Model Predictive Control (MPC) strategy, is used here to maximize the information amount from observations.And semi-physical simulation experiments were carried out to verity the A 2 CL algorithm's performance finally.Results proved that the presented A 2 CL algorithm based on information optimization strategy for MMRS cannot only enhance the CL accuracy effectively but also have good robustness.
This paper focuses on an internal model tracking control problem of an uncertain impulsive switched system. The contribution of this study is to develop an internal model tracking control law integrated with the guaranteed cost control algorithm for uncertain impulsive switched systems. Firstly, an uncertain impulsive switched system with guaranteed cost control is stated and some important theorems on stabilization and controller design are given. Meanwhile, Linear Matrix Inequality is introduced to formulate the optimal guaranteed cost control law of the uncertain impulsive switched system. Secondly, an internal model tracking control structure is presented combining with guaranteed cost control and LMI optimization. By establishing the extended state space equation, tracking control problem of uncertain impulsive switched systems can be transformed into a state stabilization problem. Finally, simulations are carried on to demonstrate the proposed algorithm.
'/%3e%3cuse xlink:href='%23a' transform='rotate(23.036 .093 25.536)'/%3e%3cuse xlink:href='%23a' transform='rotate(45.87 1.273 16.18)'/%3e%3cuse xlink:href='%23a' transform='rotate(69.945 .996 12.078)'/%3e%3cuse xlink:href='%23a' transform='rotate(20.66 -19.689 31.932)'/%3e%3c/svg%3e)
