Ishant Juyal’s approach
I would start by clarifying the goal. Siemens is trying to improve OEE across factories that still depend on old CNC machines, but we cannot replace the machines or require a full network upgrade. So the product has to be low-disruption, retrofit-friendly, and valuable even in plants with poor connectivity.
The main users I would think about are plant managers, maintenance teams, and machine operators. Plant managers care about OEE, downtime, and production loss. Maintenance teams care about knowing which machine is likely to fail and what action to take. Operators care about simple alerts and not being overloaded with another complicated system.
My solution would be a retrofit monitoring layer for legacy CNC machines. Instead of trying to integrate deeply with the old machine controller, we can use a small external edge device connected to non-invasive sensors. For example, current sensors, vibration sensors, temperature sensors, and maybe acoustic sensors can help us detect whether the machine is running, idle, overloaded, or behaving abnormally. Where possible, we can also connect to existing PLC signals, but that should not be mandatory because many machines may not expose reliable digital outputs.
The edge device would process basic signals locally, so we do not need constant high-bandwidth connectivity. It can identify machine state, capture anomalies, and send only important events or summarized telemetry to the cloud or local server. This reduces the dependency on factory-wide network upgrades. For factories with weak connectivity, the device should store data locally and sync whenever connection is available.
The product would have three main layers.
First, real-time machine monitoring. The system should show which machines are running, idle, stopped, or in fault condition. This gives teams a live view of production health.
Second, OEE dashboarding. We should calculate availability and performance using machine runtime, idle time, downtime, and production targets. Quality data may be harder to capture automatically from old CNC machines, so in V1 we may need manual input or integration with existing quality logs. The dashboard should show OEE trends by machine, line, shift, and factory.
Third, predictive maintenance alerts. Initially, I would not promise very advanced AI. I would start with rule-based and anomaly-based alerts: unusual vibration, current spikes, temperature increase, longer cycle times, or repeated stop-start patterns. Over time, as we collect more machine history and maintenance records, we can move toward better predictive models.
For MVP, I would not roll this out across all factories. I would pick one or two factories with a mix of old CNC machines and install the retrofit devices on maybe 10–20 machines. The first goal would be to prove that we can accurately detect machine state and reduce unplanned downtime. Once we prove reliability and ROI, we can scale to more machines and factories.
Success metrics would include improvement in OEE, reduction in unplanned downtime, reduction in mean time to detect issues, maintenance alert precision, and adoption by maintenance teams. I would also track false alerts because if the system creates too many noisy alerts, maintenance teams will stop trusting it.
The biggest risks are sensor accuracy, installation complexity, alert fatigue, and resistance from factory teams. To reduce these risks, the installation should be non-invasive, alerts should be explainable, and the dashboard should focus on a small number of actionable insights rather than overwhelming users with raw data.
So overall, I would position this as a lightweight industrial IoT retrofit platform for legacy machines: edge-first, low-bandwidth, non-invasive, and focused first on visibility, then on predictive maintenance.