AI + Digital Twins: Industrial Tech Gets a New Lease on Life | BlueRun Ventures Hard Tech Salon

When AI and digital twin technology "stack buffs," China's industry is facing a platform-level opportunity.

From materials innovation, R&D design, process simulation and optimization, manufacturing, to product delivery and maintenance — every critical link matters for creating high-value-added industrial products. The challenge China's industry faces today is this: on top of an already well-developed flow of goods and logistics, the industrial system needs a systematic upgrade at the data-flow level, while becoming comprehensively intelligent to better match and adapt to changes on both the supply and demand sides.

Two factors are crucial here: digitalization and intelligence. How will AI and digital twins accelerate this process?

BlueRun Ventures recently launched its Industrial Technology Salon, bringing together academia, industry, and startups to explore the opportunities emerging in industrial tech. The event was packed with substantive, hardcore content — we've extracted the most essential parts below. Enjoy!

China Will See Platform-Level Opportunities in Industry

Shi Jianping

Investment Partner, BlueRun Ventures

China's manufacturing digital transformation has mainly focused on commodity circulation, yet the industrial sector itself still largely concentrates on the mid-to-low end. But in this major cycle of global supply chain restructuring, China will increasingly focus on creating production value.

In recent years, AI has iterated rapidly and begun entering more accessible,普惠化 scenarios — it's no longer out of reach for ordinary people. I believe AI will bring tremendous value and help to human civilization and the world.

Today, artificial intelligence and high-fidelity digital twin technology are already empowering every link in industry while creating greater value. We see a major trend: industry will shift toward real-time collaboration — across departments, roles, and even industrial chains — ultimately producing platform-level opportunities.

How to Build Next-Generation Industrial Digital Twin Systems

He Zhan

Head of NVIDIA Omniverse Business, China

NVIDIA Omniverse proposes the concept of "full-fidelity digital twins." We have four-dimensional expectations for such digital twins: a unified, easily shareable expression format; physical accuracy and precision; whether AI can play a decisive role in next-generation digital twins; and serviceability in real, physical environments.

Why does industry need AI and digital twins? AI requires data at scale, with diversity and accuracy — this is impractical and difficult to obtain through traditional means, hence the need for SDG (Synthetic Data Generation). For AI trained in simulation to succeed in the real world, it must be trained in environments indistinguishable from reality — meaning it must be physically accurate and obey the laws of physics.

In summary, "AI-infused and physically accurate" industrial digital twins cover all aspects of intelligent supply chains, with many industries already applying this technology. From practical cases, we've learned some lessons: digital twins start from fully realistic visualization; the chain needs to be connected with real-time data linking and interaction.

Digital Twins Are Model-Based Systems Engineering (MBSoSE)

Xing Jun

Vice President, PERA SIM; Deputy Director, National Institute of Industrial Software and Advanced Design

PERA SIM believes that digital twins are system-level virtual-physical mappings, not one-to-one mappings. Their essence is establishing different models and combinations in digital space, with twin bodies representing business objectives; actual physical space data drives model evolution, forming description, diagnosis, prediction, and control of actual physical space and business objectives. In other words, business is the core and goal, data is the technology, models are the core software, and software is the carrier.

Digital twins need to start from a business perspective. Building mathematical models and decision optimization around customer business objectives is the core of the digital twin chain. This means facing complex, large-scale customer needs and combining with industry ecosystem IoT, foundations, industrial internet platforms, visualization, etc., to provide complete services.

Digital twin construction is a systematic engineering effort requiring full-lifecycle scenario consideration. First, it's built on comprehensive models constructed from data generated by physical entities during R&D, production, operation, and maintenance; the purpose of building it is to advance specific business functions. It should reflect real-time current conditions while also providing early warning and rehearsal for future trends.

AI in Industrial Solvers

Lin Zhouchen

Professor, Peking University; IAPR/IEEE/CSIG Fellow; National Science Fund for Distinguished Young Scholars

Early industrial solvers relied heavily on numerical computation and human knowledge. Today's improvements are closely tied to machine learning — AI provides core information to industrial solvers based on historical data.

Industrial solver types — at the application layer: geometric design, physical simulation, and process management; at the foundation layer: equation solvers (linear systems, partial differential equations) and optimization solvers (continuous, mixed, discrete/combinatorial optimization).

Why can AI play a role in solvers? Traditional industrial solvers use mathematical computation plus heuristic rules. But this isn't enough — for example, mathematical computation has inherent stability defects that cannot be overcome. AI can provide past empirical data to improve solver performance. A good AI solver is like building blocks: replacing modules still allows it to run well — meaning faster, more accurate, and more automated.

AI application in solvers currently faces several major challenges: training data generation, whether neural network design is applicable, and whether deep relationships can be mined through data.

AI + Materials R&D

Wang Xuanze

Founder & CEO, DeepMaterial

Metallic materials represent a massive market, but currently faces problems: high-end metallic materials R&D still lags abroad; industry profits concentrate in production and sales, while pain points lie in R&D; there's a large gap between production and research — lab technology cannot quickly transfer to production, and production profits cannot feed back to R&D.

How to solve this? AI-based R&D paradigms are the optimal feasible solution. First, conduct large volumes of high-quality, low-cost, high-efficiency, high-throughput experiments simultaneously, put obtained performance and characterization into models for training and prediction, then select the most likely formulations for preparation. The benefit is that capital plays a significant accelerating role; business models can also change: first clarify market demand and customer performance requirements, shortening the R&D-to-industry process and improving matching.

Second, such models can enable entirely new material R&D. Traditional models require large numbers of experts and time to extract human-comprehensible approximate physical models; but AI can use deep neural networks to skip this cycle.

Many worry that AI-based models are black boxes with poor interpretability, but I see this as precisely AI's advantage. Traditional R&D modes expand boundaries based on existing theories, step by step toward new materials — with limited performance improvement. But if AI finds a new material, then through macroscopic and microscopic characterization analyzes why this occurs, establishing new theoretical frameworks — this may become a new mode for future academic research.

AI + Generative Design Simulation

Zhang Haixi

Founder & CEO, OptFuture

Why is generative design needed? The problem with past experience-based design is strong dependence on human experience, with massive demand-to-solution modifications in between, resulting in low efficiency. The purpose of generative design is performance-oriented, more intelligent and automated optimized design.

AI has irreplaceable advantages in generative design in four areas:

1. Structural optimization design. In selecting materials, performance, cost, manufacturing processes, and aesthetics for design needs, AI can provide large volumes of design requirements in extremely short time; AI can also provide scalable design domains not limited by given ranges; deep learning and model prediction can also enable real-time simulation.

2. New/metamaterial microstructure design. AI can realize combinations of material databases, functional lattice unit libraries, and process parameters, and see their interrelationships; establish professional knowledge bases through materials genome for intelligent optimized material selection; and design microstructures meeting requirements based on material properties.

3. Data-driven design. Data-driven integrated process-structure-performance design; multi-physics driven design, digitally integrating multi-scale features and multi-type materials to achieve structural functional fusion.

4. CAE digital twins. AI can compensate for 3D model verification technology defects — in traditional methods, using sensor-collected data for CAE simulation calculations is very slow, but through AI and model libraries for prediction and interpolation calculation, responses at all positions under all working conditions can be simulated. This way, digital twins will play a greater role in product experimental verification, full-lifecycle monitoring, and maintenance.

AI + Process Simulation Optimization

Zheng Xuan

Founder & CEO, Hetu Intelligent Manufacturing

Materials are complex living organisms — just as cancer cells constantly erode other cells causing tissue damage, a microstructural defect can affect a material's macroscopic mechanical properties. Manufacturing processes are also comparable to Nüwa creating humans — a slight deviation during processing can have major supply chain impacts. Well-known examples include ASML lithography machine lenses, light sources, and ultra-high-precision micro-motion platforms, as well as aeroengine directional solidification single-crystal blades and hot isostatic pressing superalloy turbine disks — all process masterpieces.

Traditional methods for material understanding and process application are trial-and-error, bringing risks in product R&D, production, and use. We propose "data + models + AI computation = knowledge," providing optimal solutions at the process stage. Our core focus is process simulation models: first, through accumulation of fundamental material data and optimization of boundary conditions, make process simulation models as accurate as possible and feed forward to AI, making systems more agile and efficient. Meanwhile, in the process of simplifying accurate simulation physics models to efficient surrogate models, data quality improves.

In summary, materials and processes involve high uncertainty, long cycles, and directly concern product quality. We must empower enterprises through AI to solve this problem.

AI + Additive Manufacturing

Zhang Guoliang

Founder & CEO, Leimo Technology

Additive manufacturing breaks through traditional design thinking limitations, enabling more optimized structures and designs better suited to different scenarios and needs, thus largely liberating design and productivity; it requires interdisciplinary integration, has high technical barriers, and multiple linked factors collectively determine whether it can truly meet downstream application scenarios and user needs.

Selective Laser Melting is the most widely used metal additive manufacturing process. Macroscopically, laser light from the laser source passes through optical path system processing to form high-quality beams; metal powder materials melt layer by layer under beam energy forming melt pools, while fluid wind fields remove harmful substances produced during powder melting; melt pools rapidly solidify to complete metal entity forming.

Microscopically, during printing, melt pool depth, width, porosity, stress, inclusions, and accompanying harmful substances like ablation, black smoke, and spatter are all important factors affecting print quality, requiring systematic optimization through equipment, materials, and processes to control.

From design to process to manufacturing, main pain points concentrate in three areas: experience-driven difficulty in standardization, discrete data difficulty in automation, and static processes difficulty in intelligence. By incorporating AI capabilities at each critical link, first achieving local optimal solutions — such as intelligent design, intelligent process simulation and optimization, adaptive manufacturing process control, etc. — then combining with digital twin capabilities to connect all links forming global optimal solutions.

Although additive manufacturing's design, process, equipment, and materials have strong coupling, its inherently higher degree of digitalization also gives it broader prospects in combining AI and digital twins. Whether in materials R&D, intelligent design simulation, or process optimization and process quality control, Leimo Technology is actively exploring, hoping to drive additive manufacturing toward true intelligent manufacturing through "AI + digital twin."

Digital Twins and Collaborative Intelligent Manufacturing

Gong Minyan

Founder & CEO, Ziqian Technology

The problem with AI + digital twin applications in industry over the past 30 years: the digitalization industry used know-how to create software for each link, producing massive amounts of software, resulting in poor business consistency, high integration costs, and difficult data sharing. Previous solutions were process-segment integration, but this approach had high integration and maintenance costs and was slow.

One possible direction for digital twin foundations is a semantic web + implicit AI model-based framework. The essence of semantic web is reorganizing data through normalized paradigms to form data interconnections, which can constrain implicit models.

But building and using enterprise system digital twins remains a technical challenge today, because no ready-made integration technology can build complete enterprise models. Four aspects need improvement: industrial-oriented toolchains; explicit expression, storage, and migration of knowledge; integration of massive heterogeneous data; and building capabilities combined with implicit AI models.

Ziqian Technology's attempt is establishing an architecture called UEOS. By building a simplified semantic web-like model, it aggregates data from different value chains and sends data to artificial intelligence, thereby providing faster constraints and suggestions. In summary, to form true digital twins, we must transform from the blind men touching the elephant state to a paradigm where different perspectives extract content from the same ontological model.

** ▶ More Exciting Content ◀**

Originating in Silicon Valley, BlueRun Ventures was established in 2005 and is a venture capital firm focused on early-stage startups.

Currently, BlueRun Ventures manages multiple USD and RMB dual-currency funds in China, with assets under management exceeding RMB 15 billion, making it one of the largest early-stage funds domestically. Its investment stages concentrate on Pre-A and Series A, covering hard tech and innovative interaction, enterprise technology, new consumption, and healthcare. It has cumulatively invested in over 150 startups, including Li Auto, Waterdrop, QingCloud, Guazi.com, Qudian, Songguo Mobility, Ganji.com, Monster Charging, Yuntu Semiconductor, Machenike, Clouds Intelligence, Anxin Wangdun, BioMap, and others.

BlueRun Ventures has been ranked first in Zero2IPO's "China Top 30 Early-Stage Investment Institutions" and ChinaVenture's "China Best Early-Stage Venture Capital Institutions TOP30," and was named among Preqin's Top 10 VC Fund Managers Globally for Sustained High Returns.

Additionally, BlueRun Ventures has consecutively received honors from Forbes China, 36Kr, Cyzone, Caixin Media, CBNweekly, Jiemian, and other media institutions, including "China's Best Early-Stage Institution of the Year," "China's Top Venture Capital Firm," "Most Entrepreneurur-Friendly Early-Stage Institution of the Year," and "Most Influential Early-Stage Institution of the Year."