Make the machines you already own tell you what they are doing.
Retrofitting sensing and analysis onto existing industrial equipment: condition monitoring, predictive maintenance, vision inspection and integration with the systems you already run.
Start with the machine you already have
The most practical entry point is rarely a new machine. It is the equipment already installed, already running, and already central to your output. Retrofitting sensing onto it is lower risk than replacing it, proves value before capital is committed, and does not require the production line to change while the question is still open.
Much of the useful information is available from outside the machine, without modifying it or voiding anything.
What can be sensed without opening the machine
- Vibration. Bearing condition, imbalance, misalignment and looseness usually appear here first.
- Current and power. Motor load, cycle timing, idle time and developing mechanical resistance, measured at the supply with a clamp rather than inside the panel.
- Temperature. Surface and bearing temperatures, and their trend relative to load.
- Acoustic signature. Changes in running sound, including ultrasonic ranges, which can indicate leaks and early mechanical wear.
- Cycle and utilisation. How often the machine actually runs, how long each cycle takes, and where time is lost.
Condition monitoring before prediction
“Predictive maintenance” is what most people ask for. It is worth understanding what it requires.
Predicting a specific failure with a useful lead time generally needs recorded examples of that failure developing. Most operations do not have those, because failures are rare and historically nobody was recording the right signals when they happened. Starting a project by promising prediction from data that does not exist is how these programmes lose credibility.
The realistic sequence is condition monitoring first: establish what normal looks like for each machine, detect departures from it, and alert on those. This is genuinely useful on its own, it catches a good proportion of developing faults, and it builds the labelled history that makes real prediction possible later. As failures occur and are recorded against the signals that preceded them, the dataset accumulates and the prediction question can be revisited with evidence behind it.
Applications
- Condition monitoring and anomaly alerting. Per-machine baselines and departures from them.
- Developing predictive capability. Built on accumulated history rather than promised on day one.
- Visual inspection. Camera-based checking for defects, presence, orientation or counting, where consistency matters more than human judgement.
- Energy and efficiency. Consumption per unit produced, idle draw, and which assets and shifts drive the cost.
- Utilisation and downtime analysis. Measured machine availability and where production time is actually going, rather than what the schedule assumes.
Integrating with PLC, SCADA and your existing systems
A retrofit system that cannot talk to anything is another island. Data usually needs to reach your existing SCADA, historian, maintenance system or ERP, and sometimes needs to read from the PLC rather than only sensing externally. Protocol choice, read-only versus read-write access, network segmentation and behaviour when the other system is unavailable are settled at design time.
Where operational technology and IT networks meet, segmentation and access control are part of the design rather than an afterthought. Raise your site’s requirements early, because they constrain the architecture.
Where this must not sit
Analysis of this kind is advisory. It informs maintenance planning and operational decisions. It is not a substitute for a machine’s safety systems, interlocks or protective functions, and it should not be placed in a position where a wrong inference can cause harm. Safety functions have their own engineering discipline and their own approval requirements. If your application involves safety-related control, say so at the start so the boundary is drawn correctly and explicitly.
A realistic first project
Instrument a small number of representative machines. Collect properly for a defined period. Establish baselines and useful alerting. Review what the data actually showed and what it is worth. Then decide whether to extend across the site, with the decision supported by evidence from your own equipment rather than a vendor’s case study.
Discuss your equipment
Tell us what the machines are, what failures or losses hurt most, and what you already measure. Related: AI-enabled IoT development, industrial IoT and equipment monitoring and cloud and dashboards.
Tell us about your equipment
Describe the machines, the failures or losses that hurt most, and what you already measure.
What clients ask before starting
Do we have to modify our machines?
Usually not. A great deal can be measured from outside: vibration, supply current and power, surface and bearing temperature, acoustic signature, and cycle timing. Retrofitting external sensing avoids disturbing equipment that is working.
Can you actually predict failures?
Predicting a specific failure with useful lead time generally requires recorded examples of that failure developing, which most operations do not yet have. The realistic sequence is condition monitoring first, which is useful on its own and builds the history that makes prediction possible later.
Will it integrate with our SCADA or maintenance system?
That is the intention, since a retrofit system that cannot pass data to anything becomes another island. Interfaces, access level and network segmentation between operational and IT networks are settled at design time.
Can this control the machine or replace our safety systems?
No. Analysis of this kind is advisory and informs maintenance and operational decisions. Safety functions, interlocks and protective systems have their own engineering discipline and approval requirements, and that boundary is drawn explicitly at the start.
How many machines should a first project cover?
A small number of representative machines, instrumented properly and collected over a defined period. That produces evidence from your own equipment on which to base a decision about extending across the site.