Bullen Ultrasonics is turning roughly 2 billion manufacturing data points into a test bed for industrial AI, partnering with Phenx on a machine-learning initiative designed to reduce process variation and improve production consistency. The Ohio manufacturer has received a $23,100 grant through the Ohio Smart Manufacturing Program, funding an early phase of an Autonomous Process Optimization project that could offer a practical model for how smaller manufacturers deploy AI without handing control of production to an autonomous system.
For manufacturers, the harder AI problem is increasingly not generating text or images. It is figuring out how to make a physical production process more predictable without disrupting the machines and people that already keep a factory running.
That is the problem Bullen Ultrasonics is tackling.
The precision-machining company has received a $23,100 grant through the Ohio Smart Manufacturing Program to advance an Autonomous Process Optimization initiative with industrial AI engineering company Phenx. The project applies machine learning to production data generated by Bullen’s customized equipment, with the goal of identifying process adjustments that can improve stability, efficiency and consistency.
Bullen says the project will analyze approximately 2 billion data points collected from its manufacturing equipment. Rather than immediately connecting an AI model to production machinery, the companies are first using historical data to build and validate a digital representation of the targeted manufacturing process.
That distinction matters.
A digital twin is a virtual representation of a physical asset or process that can incorporate operational data and support simulation, prediction and optimization. In manufacturing, it can provide a safer environment for testing changes before they are introduced to the factory floor. McKinsey has identified digital twins as an important mechanism for using AI to simulate manufacturing scenarios and optimize processes before making real-world changes.
In Bullen’s case, Phenx has developed a digital twin using actual production data. The current phase involves validating the optimization algorithm inside that virtual environment. If the results hold up, Bullen plans to pilot the algorithm on one machine and conduct an extended production prove-out.
The approach reflects a broader shift in industrial AI: companies are moving away from broad promises of “smart factories” and toward narrowly defined use cases where improvements can be measured against operational KPIs.
Bullen’s potential targets include reducing process variation, shortening production cycles and making manufacturing performance more predictable. Those are conventional manufacturing objectives, but machine learning can examine relationships across large volumes of operational data that would be difficult for engineers to isolate manually.
For Bullen, the data foundation appears to be as important as the AI itself. The company says it has spent several years collecting, organizing and understanding equipment data. That preparation gives the project something many industrial AI initiatives lack: a substantial historical dataset tied to a specific production environment.
The challenge now is converting that data into a reliable decision-support system.
Phenx founder and CEO Saurabh Sarkar said the company’s work with Bullen combines machine learning and modeling with the manufacturer’s existing knowledge of its machines and processes. The digital twin also provides a way to test optimization strategies before they affect live production.
That is a useful model for enterprise AI adoption. The objective is not simply to predict what a machine will do. It is to determine whether an algorithm can make a recommendation that is accurate enough, explainable enough and operationally safe enough for engineers to trust.
This is where Bullen’s project differs from the consumer-facing AI boom around large language models such as those developed by Google, Microsoft and other major technology companies. The value proposition here does not depend on a general-purpose AI assistant. It depends on specialized models trained against industrial data and constrained by a specific manufacturing objective.
Gartner’s research reinforces the importance of that distinction. Its 2025 research found that only 41% of generative AI prototypes and 42% of nongenerative AI prototypes reached production, highlighting the gap between experimentation and operational deployment.
Manufacturing organizations face an additional hurdle: production systems cannot tolerate the same level of experimentation as software applications. An inaccurate recommendation can affect yield, quality, equipment life or worker safety.
That makes simulation, validation and human oversight central to industrial AI.
Bullen says its broader strategy is to use AI to support human decision-making rather than replace its engineers, machinists and manufacturing specialists. If the initial project succeeds, the company could apply the same methodology to other manufacturing processes.
The ecosystem around the project is also notable. The Ohio Smart Manufacturing Program is designed to help small and medium-sized manufacturers adopt advanced digital technologies, while the University of Dayton Research Institute supported Bullen’s participation through technical discovery and project preparation.
For smaller manufacturers, that funding model may be as significant as the technology itself. Large industrial organizations can afford dedicated data science, automation and digital-twin teams. Smaller companies often have the operational expertise and machine data but lack the resources to translate those assets into AI systems.
Bullen’s initiative suggests a more incremental path: identify one high-value process, use historical data, build a simulation environment, validate the model and only then move toward production.
That approach is increasingly relevant as manufacturers look beyond AI pilots. IDC says the next phase of manufacturing AI will involve making sense of data from connected assets, expanding cloud platforms and applications, and augmenting the industrial workforce.
The immediate question for Bullen is whether its algorithm can deliver measurable improvements on a live machine. The longer-term opportunity is bigger: proving that a data-rich manufacturer can turn its existing operational history into a reusable AI optimization framework.
If it works, the most important output may not be an autonomous machine. It may be a repeatable blueprint for putting industrial AI into production without taking humans out of the loop.
Market Landscape
Industrial AI is entering a more pragmatic phase. Manufacturers are prioritizing predictive maintenance, quality control, scheduling and process optimization over experimental AI applications that lack a clear connection to operating metrics.
McKinsey’s 2025 research found that 90% of technology use cases among the latest Global Lighthouse Network manufacturing sites incorporated AI, while roughly two-thirds of surveyed manufacturing COOs remained at the exploration or targeted-implementation stage. Only 2% reported AI as fully embedded across all operations.
That gap creates an opening for specialized industrial AI platforms and engineering firms such as Phenx. It also explains why digital twins are gaining importance: they can create a controlled environment in which AI recommendations can be tested before production deployment.
The competitive landscape extends from specialist industrial AI providers to major technology ecosystems. Microsoft, Amazon, Google, NVIDIA, Siemens, Rockwell Automation and other industrial and cloud technology vendors are building infrastructure around industrial analytics, simulation, edge computing and AI.
For enterprise manufacturing teams, the key differentiator is unlikely to be access to a generic AI model. It will be integration with machine data, manufacturing execution systems, controls, sensors and existing engineering workflows.
Bullen’s project therefore represents a relatively small financial investment but a strategically important category of deployment: AI-driven process optimization tied directly to physical production.
Top Insights
- Bullen Ultrasonics is applying machine learning to roughly 2 billion manufacturing data points, targeting process stability, efficiency and consistency in precision machining operations.
- Phenx has built a digital twin that lets engineers validate optimization algorithms virtually before introducing algorithmic recommendations to live manufacturing equipment.
- The project illustrates a broader enterprise AI shift toward narrowly defined, measurable industrial use cases rather than open-ended experimentation with generative AI.
- Small and medium-sized manufacturers could benefit from similar grant-supported approaches that combine existing operational data, engineering expertise, digital twins and machine learning.
- Enterprise adoption will depend on validation, human oversight and integration with production systems as much as model accuracy or computing infrastructure.
Power Tomorrow’s Intelligence — Build It with TechEdgeAI
