Case Studies
Our Multi-Sigma AI software helps accelerate R&D at large corporations you admire, small organizations you’re hearing buzz about, and academic institutions you respect. Since our software creates a competitive advantage, most of our relationships and projects involve NDAs (non-disclosure agreements).
The following case studies are the precious few we have permission to share. We hope they offer a glimpse into the versatility of our software and the capabilities of our team.
Want a clearer picture for your use case? Please don’t hesitate to contact us.
Artificial Heart
AIST / National Institute of Advanced Industrial Science and Technology (Japan)
At the National Institute of Advanced Industrial Science and Technology (AIST), scientists are working on an exciting project to create an artificial heart. They’re using special bearings and smart planning with artificial intelligence to make the artificial heart work better. The goal is to improve both the power of the bearings and how safe the artificial heart is for our blood cells. The team wants to help people who have to wait a long time for a new heart. By combining different areas of expertise, they’re trying to change the way we do heart transplants and make artificial hearts a more hopeful option for people with heart problems.
Challenges
Heart diseases are a major cause of death in Japan, and the prolonged waiting period for heart transplants necessitates the development of an artificial heart using hydrodynamic bearings. The design of hydrodynamic bearings involves numerous design parameters and optimizing them through trial and error has its limitations.
There are 7,200 combinations of input conditions, including groove number (3-18, 16 options), groove angle (10-180 degrees, 18 options), groove entrance depth (0.05-0.25 mm, 5 options), and groove exit depth (0.05-0.25 mm, 5 options).
Solutions
Using an innovative experimental planning method leveraging AI, we conducted an analysis combining neural networks and multi-objective genetic algorithms. Based on a minimal set of 30 to 60 simulation experiments, we optimized complex designs with multiple inputs and objectives. Using AI analysis through Multi-Sigma from Aizoth Co., Ltd. (Headquarters: Tsukuba, Ibaraki), we achieved efficient analyses, showcasing the power of AI in optimizing intricate designs.
Results
We used a neural network to create an AI predictive model and then employed a multi-objective genetic algorithm to find the optimal design for hydrodynamic bearings. Our goal was to simultaneously increase generating force and reduce red blood cell damage. Contrary to conventional thinking that more grooves are better, our AI-driven factor analysis revealed the crucial role of groove depth, challenging traditional design guidelines. Comparing the new designs to conventional ones showed potential improvements in both generating force and reducing red blood cell damage. Despite the usual belief, our results suggested that designs with greater groove depth could be more effective. Through analyses based on 30 to 60 iterations for each of the 7,200 combinations, we efficiently optimized the design, achieving innovative solutions with less than 1/100th of the typical effort, showcasing the transformative impact of AI in discovering efficient designs.
Aluminum Recycling
NEDO / New Energy and Industrial Technology Development Organization
Aizoth was chosen for the “Aluminum Material Advanced Resource Circulation System Construction Project” implemented by the New Energy and Industrial Technology Development Organization (NEDO) under the theme of “Development of Upgrade Recycling Technology for Aluminum Resources Toward the Construction of a Resource Circulating Society.” The project aims to develop upgrade recycling technology to regenerate aluminum resources and produce recycled aluminum with high functionality.
Aluminum is known for its lightweight, corrosion resistance, easy processing, and high thermal conductivity, making it a sought-after material for lightweight applications such as automobiles. However, in the current recycling processes, impurity concentrations increase with each recycling cycle, leading to a decrease in the material’s strength, ductility, and other functions, ultimately resulting in landfill disposal. Additionally, while some aluminum is recycled, the primary metal for high-purity extrusion materials is consistently reliant on imports from overseas.
Challenges
From experimental data, it was necessary to estimate the cost and environmental impact for each experimental condition and optimize the manufacturing conditions in a multi-objective manner along with the objective variables related to functionality. Only 18 pieces of experimental data were available. From this, the optimization of conditions of a complex process was needed, with six parameters for manufacturing conditions and six parameters targeted for objectives.
Solutions
A practical Life Cycle Assessment (LCA) evaluation was conducted, considering factors like scale effects, to estimate costs and environmental impacts for each experimental condition. The AI was trained on 18 data points encompassing six experimental conditions within the aluminum upgrade recycling process, specifically in the heat treatment stage, including impurity concentration, temperature, time, and pressure. Experimental results included material properties such as hardness, tensile strength, and yield strength, along with six objective variables related to greenhouse gas emissions and cost. Utilizing the Multi-Sigma auto-tuning technology, the AI model was trained to create a high-precision neural network prediction model while suppressing overfitting. The resulting prediction model, coupled with a multi-objective genetic algorithm, was then employed to explore optimal processing conditions that maximize metal functionality while minimizing greenhouse gas emissions and cost.
Results
From just 18 sets of experimental data, we successfully explored manufacturing conditions that not only result in greenhouse gas emissions of 1/10 or less compared to primary metal production but also keep costs in check. Furthermore, these conditions achieve functionality comparable to high-tensile steel sheets.
Harvest Predictions
KAGOME AGRIFRESH CO, LTD.
AIZOTH collaborates with KAGOME AGRIFRESH CO., LTD. , to develop an AI-powered fresh tomato yield prediction system. The primary aim is to enhance the accuracy of yield predictions, crucial information for supply-demand balancing in fresh tomato production. KAGOME implemented this system in large-scale farms cultivating KAGOME brand fresh tomatoes since February 2022. By leveraging AI trained on historical weekly reports accumulated from KAGOME’s contracted farms, KAGOME has achieved the remarkable feat of predicting tomato harvest quantities up to 5 weeks in advance.
Challenges
KAGOME, involved in the sale of processed tomato products and fresh tomatoes, encountered significant challenges in accurately forecasting fresh tomato harvest quantities. These predictions were pivotal for the company, given the multifaceted nature of variables impacting fresh tomato yields, including climate and cultivation methods. Previously, relying on the intuition and experience of farm shipment personnel to forecast fresh tomato harvest quantities proved challenging due to the complexity of these influencing factors.
Solutions
Initially exploring the development of an in-house meteorological forecasting system to improve tomato harvest quantity predictions, KAGOME faced substantial challenges in creating such a system. Consequently, attention turned toward the wealth of data acquired from farm records. Each weekly report meticulously detailed over 100 items per tomato variety, encompassing crucial parameters such as temperature, humidity, watering schedules, and actual harvest yields. This data was utilized for AI training. Recognizing that training AI models with all available features yields higher prediction accuracy compared to manually extracting specific features, KAGOME inputted all data to create a predictive model.
Results
The strategic application of AI for predictions has enabled KAGOME to forecast tomato harvest quantities up to 5 weeks in advance.
Medical Care for Patients with Chronic Lower Back Pain
Multi-Sigma® uses patient data to predict pain relief and highlight the factors that matter most, then recommends a personalized mix of therapies to treat chronic low back pain. PDF
Optimizing Sleep Conditions through AI Chain Analysis
Multi-Sigma® links multiple AI models to study how sleep, caffeine, exercise, and environment affect fatigue and focus. It predicts outcomes and recommends the best conditions to improve sleep quality and next-day performance. PDF
Pharmaceutical Development using Multi-Sigma
Multi-Sigma® is used to predict drug activity and side effects for Alzheimer’s drug candidates, then balance both through multi-objective optimization. By linking AI models and analyzing molecular descriptors, it proposes solutions that boost drug effectiveness while reducing the risk of vascular disorders. PDF
Hydration Free Energy Prediction in Molecular Design
Multi-Sigma® is applied to predict and optimize hydration free energy, a key property in drug discovery. By analyzing molecular descriptors, it identifies factors that influence results and suggests conditions that minimize hydration free energy to support faster, more efficient molecular design. PDF
Balancing Density Control / CO₂ Absorption in MOF Synthesis
Multi-Sigma® is applied to MOF synthesis to balance material density with CO₂ adsorption capacity. By linking AI models and running multi-objective optimization, it identifies synthesis conditions that deliver stable density and strong CO₂ capture performance. PDF
Multi-Objective Optimization of Automotive Fuel Efficiency / Engine Power
Multi-Sigma® is used to model and optimize the trade-off between fuel efficiency and engine power. By analyzing key factors like weight, displacement, and acceleration, it predicts performance and proposes parameter sets that balance efficiency with strong output. PDF
Exploration of next-generation Solar Cell Materials
Multi-Sigma® is applied to explore new solar cell materials by predicting formation energy and bandgap from material descriptors. Through factor analysis and multi-objective optimization, it identifies compounds with stable structures and ideal light absorption properties for next-generation solar cells. PDF
Optimizing the Aluminum Recycling Upgrade Process
Multi-Sigma® is used to optimize aluminum recycling by improving strength and durability while cutting greenhouse gas emissions and costs. Through factor analysis and multi-objective optimization, it identifies key process parameters and proposes the best conditions for sustainable, high-performance recycling. PDF
Designing an Artificial Heart
Multi-Sigma® was applied to the design of an artificial heart pump, where it built a surrogate model from only 50 simulations instead of 7,200. The AI predicted thrust force and hemolysis, identified groove depth as the key factor, and optimized parameters to maximize pump performance while minimizing blood cell damage. PDF
Risk Prediction and Factor Analysis of Alzheimer’s Disease
Multi-Sigma® was used to build a neural network that predicts Alzheimer’s disease risk with 96% accuracy. Contribution analysis showed that CDR was the strongest predictor, while factors like brain volume, education, and MMSE scores also played key roles in shaping risk. PDF
Analysis of Retail Sales Time Series Data
Multi-Sigma® was applied to retail sales forecasting, building a neural network that predicts sales one week ahead. Contribution analysis revealed the strongest drivers, such as seasonal patterns, prior week sales, and discount campaigns, giving insights into department-level performance. PDF