Artificial Intelligence and the Internet of Things offer opportunities to enhance industrial monitoring, asset visibility, maintenance and decision-making in operational environments. However, while a working prototype can help illustrate the general value of a given technology, turning an idea into a production-ready system requires a keen focus on factors such as data quality, system integration, deployment constraints and commercial viability.
Here are 5 considerations for teams building industrial technology ventures.
- Define the operational problem first
One common challenge for teams developing technology solutions is the tendency to conflate the problem and solution, often with undesirable consequences. If a manufacturing facility experiences frequent unplanned equipment downtime, for example, a team developing a predictive maintenance system should ask themselves questions such as:
• What equipment failures are we trying to predict?
• What sensor measurements or historical records do we have at our disposal?
• What maintenance decisions are being made currently, and what data influences these decisions?
• What information would influence a decision or maintenance schedule change for operators or technicians?
These questions can provide clarity on whether an AI-based approach is appropriate, and allow developers to design a system that delivers tangible value to a given audience.
A clearly defined use case also helps set expectations, and measure whether a given solution meets its intended goals.
- Treat data quality as a design requirement
At the heart of most AI systems are analytics models, which are typically built using training data. However, data requirements can differ considerably from one application area and use case to another, and industrial environments often present significant data-related challenges.
For example, a manufacturing facility may have a variety of PLCs, MES, IoT sensors, ERP systems and other technologies that can store data in various formats, with inconsistent timestamps or identifiers. Developers should therefore spend considerable time evaluating data-related requirements, including:
• Completeness of the data
• Timestamp differences
• Possible missing or repeated values
• Equipment identifiers
• Historical data availability
• Access and security requirements
Having well-documented design requirements can help identify any gaps beforehand.
- Design for integration
Industrial technologies rarely exist in a vacuum, and developers should recognize that their solution will typically need to interface with other technologies and environments. A condition monitoring application, for instance, may need to retrieve equipment data, associate measurements with equipment tags, and visualize this information within a dashboard or the target environment.
The integration architecture should therefore be considered during the design process, and developers should strive to understand how the data flow, API keys and other elements might fit into the target environment. Particular attention should be paid to system compatibility, failure scenarios, authentication and access control, and recovery procedures.
By recognizing the requirements at an early stage, developers can minimize the technical and commercial risks associated with such integration requirements.
- Separate monitoring from prediction
Operational data analytics and predictive maintenance use cases share some similarities, but developers should treat them as separate functions. An operational monitoring system, for instance, may simply raise alerts when a given temperature threshold is passed, whereas a predictive maintenance model would attempt to predict the likelihood of a failure based on relevant information and historical trends.
The latter typically requires more extensive data preparation, feature engineering, evaluation and testing before launch. It can therefore be useful to release a monitoring system first, establish stability and gather evidence of value, and later transition to a predictive maintenance model if the evidence supports it.
This approach can reduce unwarranted complexity, and help developers avoid analysis paralysis by only pursuing additional capabilities if the existing system demonstrated value.
- Validate technical and commercial feasibility together
A production environment is rarely a technical showcase, and developers should recognize the importance of building solutions that are viable in the real world.
Industrial buyers typically seek to understand the costs, benefits and risks associated with adopting a given technology, and a pilot should therefore go beyond an evaluation of technical capabilities. In addition to assessing the technical requirements and opportunities for improvement, developers should therefore seek to understand whether the solution provides value in the context of a given business.
For industrial AIoT ventures, this means balancing customer discovery and technical validation throughout the pilot environment. These 5 points are part of a wider set of venture-building considerations, and
Aperture Venture Studio offers additional resources on the intersection between technology and venture building.
Conclusion
Developing industrial AIoT products presents unique challenges, and a strong focus on areas such as data quality, system integration and value proposition can help teams avoid some common pitfalls.
By focusing on the problem first, validating technical and commercial requirements and only implementing core analytics functionality when the time is right, developers can ensure that they pursue realistic use cases that have a viable path to launch and long-term adoption.
Discussion: For developers of industrial IoT or AI applications, which challenge tends to require more effort in practice: equipment or data integration requirements, edge computing design, or demonstrating value in a given environment?
Top comments (0)