Reducing Equipment Downtime Across Multi Plant Manufacturing Operations
How We Helped a Heavy Manufacturing Company Empower Field Technicians with AI Diagnostics Online and Offline
Client Snapshot
| Industry | Region | Scale | Operations | Key Stakeholders |
|---|---|---|---|---|
|
Steel Fabrication & Heavy Manufacturing |
Middle East (Saudi Arabia) |
400+ employees operating across 3 production plants |
24/7 production environment with 65+ machines including CNC equipment, |
VP of Operations, Maintenance Manager, Plant Supervisors |
The Challenge
This manufacturer was facing a critical operational bottleneck. Their team of 18 maintenance technicians was responsible for keeping over 65 machines running across three geographically separated plants. When equipment broke down, junior technicians lacked the experience to diagnose issues independently and had to call senior engineers for guidance, often waiting 30 to 45 minutes for a callback during peak hours.
The problem was compounded by connectivity limitations. One of their plants housed heat treatment furnaces in a partially underground facility where mobile and Wi Fi signals were virtually non existent. Technicians working in this plant had no way to access digital manuals, look up fault codes, or get remote support. They were entirely dependent on memory and printed binders that were often outdated.
The financial impact was significant. Unplanned downtime was costing an estimated $15,000 per hour in lost production. With three to four breakdowns occurring per week, the annual cost of slow diagnostics and delayed repairs was running into millions. Meanwhile, junior technicians were growing frustrated, and senior engineers were being pulled away from strategic work to answer routine troubleshooting calls.
Our Approach
CubePeaks deployed an AI powered field diagnostic assistant across all three plants. The solution was designed as a hybrid mobile application built with Flutter, giving it native performance on both Android and iOS devices while being maintainable from a single codebase.
Building the Knowledge Base
The first step was converting every equipment manual, service bulletin, and troubleshooting guide into searchable vector databases. These databases, packaged as downloadable files, could be browsed, selected, and stored directly on each technician’s device through a built in manual management system. Technicians could see which manuals were available, download the ones relevant to their plant, and manage storage as needed. This meant the collective knowledge of every machine on the floor was accessible from any technician’s phone.
Dual Mode AI: Online and Offline
The core innovation was the app’s dual mode architecture. In areas with connectivity, such as the main production floor, offices, and control rooms, the app operated in online mode, streaming detailed diagnostic responses from a cloud hosted backend. These cloud responses were richer, included direct citations to manual sections, and leveraged more powerful AI models for complex troubleshooting scenarios.
When a technician walked into the underground plant or any zone with poor signal, the app detected the connectivity change and seamlessly switched to offline mode. In this mode, an on device AI model took over, running locally on the phone with full function calling capabilities. The local AI could search through the downloaded manual databases using agentic RAG (Retrieval Augmented Generation), meaning it did not just match keywords, it understood the technician’s question semantically and retrieved the most relevant diagnostic procedures. The transition between modes was completely transparent. Technicians did not have to toggle any settings or even notice the switch happening.
Image Based Diagnostics
Many diagnostic situations are inherently visual, such as a worn bearing, a discolored connector, or a fault code on a display panel. The app supported image diagnostics, allowing technicians to photograph equipment issues, with up to 10 images per session, and receive AI powered visual analysis. The AI could identify component wear patterns, read fault codes from display panels, and match visual symptoms to known failure modes in the manuals. This was particularly valuable for junior technicians who could not yet recognize visual cues that experienced engineers spotted instantly.
Hands Free Voice Input
Field technicians frequently have their hands occupied, holding a flashlight, supporting a panel, or operating tools. The app included voice input capabilities through speech to text, enabling fully hands free operation. A technician could speak their question naturally, and the app would process it exactly as if they had typed it. This was critical for practical adoption on the shop floor, where typing on a phone screen was often impractical.
Configurable AI Settings
Different diagnostic scenarios required different AI behaviors. For straightforward fault code lookups, quick and precise responses were ideal. For complex, multi step troubleshooting, more exploratory and detailed responses were needed. The app provided configurable AI settings, including temperature, top K, top P, maximum token length, and the choice between GPU and CPU processing backends, allowing the maintenance manager to tune the AI behavior for different use cases and device capabilities.
Persistent Chat History
Every diagnostic session was stored in a local SQLite database on the device, including the full conversation, any images shared, and the AI’s responses. Technicians could reference past sessions when encountering similar issues, and supervisors could review diagnostic history to identify recurring equipment problems. This created an institutional memory that improved over time as more diagnostic sessions were logged.
Projected Results
| Metric | Projected Impact |
|---|---|
|
45% ↓ |
Reduction in average equipment downtime per incident through faster AI guided diagnostics. |
|
60% ↓ |
Fewer escalations to senior engineers, freeing them for strategic maintenance planning |
|
$2.1M |
Estimated annual savings from reduced downtime, fewer repeat visits, and faster repair cycles |
|
3x |
Faster first response time for junior technicians encountering unfamiliar equipment faults. |
|
100% |
Diagnostic coverage across all facilities, including zero connectivity zones. |
|
30 sec |
Average mode switching time between online and offline, completely transparent to the user. |
Heavy Manufacturing Company
“Our junior technicians used to freeze when a machine went down in the underground plant because they couldn’t reach anyone for help. Now they pull out their phone and the app walks them through it step by step, even with no signal. When they walk back to the main floor, it switches to the cloud automatically and they don’t even notice. It’s completely changed how our maintenance team operates.”