A Glimpse into the Challenges of Industrial Maintenance

Charlie is a maintenance technician working for a well-known US plastics manufacturer. Today, their high-speed automatic guillotine ground to a halt, disrupting their entire production line. With the cost of downtime growing by the minute, Charlie was sent out immediately to diagnose the issue.

To rectify the issue effectively, Charlie knows he needs to consult the repair guides and look through work order history for this asset. He opens his laptop to search for these crucial documents in the company's document management system. The complicated user interface is the first obstacle. The slow login times add to the frustration, wasting precious minutes. Once logged in, Charlie finds himself lost in an incomprehensible information architecture. The search function, far from helpful, throws up irrelevant results or outdated documents.

John takes a total of 35 minutes to find only a general maintenance guide — far from the specific documentation he urgently needed. Left with no other choice, he proceeds with the repair work. While he manages to get the machine running, the time taken was excessive and the quality of repair remains uncertain. It turns out the necessary documentation did exist, but it was buried in the depths of the corporate digital storage, isolated and therefore useless.

The true cost of this incident extends beyond the wasted time or prolonged downtime of a crucial asset. It negatively impacts overall employee morale, fails to capture important information about the nature and cause of the fault, and will likely result in a similar fault occurring again in the near future.

If only there was a better way…

Why now?

The timing couldn't be better for adopting Conversational Search in industrial settings. A confluence of technological advancements have emerged that together can solve these long-standing challenges:

• NLU Advances: Advances in Natural Language Understanding (NLU) powered by Large Language Models (LLMs) like GPT4 mean that computers can now comprehend user intent, far exceeding the capabilities of basic keyword matching algorithms.
• Digital Transformation: Company-wide digitalization of documents and workflows enables easier access to vital information.
• Mobile Ubiquity: Mobile devices are now common, offering powerful computing capabilities in the field.
• Speech-to-Text Accuracy: Advances in speech recognition allow for precise voice transcription even in noisy industrial environments.

How does this work?

Let's dig deeper into the engine that powers these revolutionary systems: the Large Language Model (LLM). The LLM empowers the AI system with a deep semantic understanding, enabling it to fully grasp the intent behind a user's words.

The LLM works by taking in a large contextual dataset and using it to make highly accurate next-word predictions. This capability is not just limited to conversational chat applications; it can also be used to enable searching through large datasets of digitized documents. A specialized variation of LLM technology can understand the semantic meaning of a conversational query and find relevant documents, even if the none of the specific keywords from the query appear in those documents. The LLM matches against the gist of what is written, not the content. Once the LLM has found the relevant documents, it can answer a worker’s question based on that information. The worker can even ask follow-up questions to further refine the answer. These answers are based on specific documentation and therefore are accurate and do not suffer from the hallucinations that commonly occur in generalised LLM implementations.

By leveraging the comprehensive capabilities of LLMs, we're not merely making incremental improvements to the workflow. Instead, we're changing the way the workforce accesses information: getting immediate insight from a question, and tracing that insight back to documentation deep within the systems of record.

Challenges Remain

While LLM-driven solutions are revolutionary and open the door to a host of new capabilities, they are not a cure-all. Several challenges and limitations still need to be addressed.

• Human Element: While AI systems like LLMs are becoming increasingly sophisticated, they're designed to augment human decision-making, not replace it. Even with the most advanced AI, human oversight is crucial for correcting potential errors and making nuanced judgments that a machine can't.
• Inconsistent Responses: Slight variations in the way a question is asked results in the LLM generating different wording for the answer. These inconsistencies can lead to technicians receiving varied guidance for the same issue, making it difficult to ensure a standards are followed consistently.
• Network Dependency: LLMs and the document repositories they integrate with are too large to run on-device and therefore robust internet connectivity is required to unlock many of their benefits.
• Legal and Compliance: The advances in LLM technologies are so recent that it is often difficult for corporate legal departments to understand how to approach legal, privacy and regulatory compliance issues resulting from the new technology.
• Data and Integration: The power of AI is largely dependent on the availability and quality of digitized data. If documents aren't digitized or integrated well with company backend systems, even the best AI solutions can't perform optimally. Proper data infrastructure is essential for maximizing the benefits of AI-driven systems.

How Datch solves this

At Datch, we have embedded Conversational Search in our tool for frontline workers. This enables technicians and engineers to quickly get reliable answers from source document, and it equips the frontline with the full knowledge of the enterprise.

Ready to see Datch in action for yourself? Schedule a demo with us.

‍

‍