Product update ·
The Bluestream PdM Platform: where it stands today, and the roadmap ahead
As of August 6th, 2026, the platform runs its loop end to end: a signal comes in, is trended, and is projected in a straight line to its alarm limit, so an alert can open with lead time to plan instead of at the limit. On that date the signals were simulated, and the work order into Microsoft Dynamics 365 was designed, not yet delivered.
Predictive maintenance only earns its keep when a health trend becomes a scheduled work order, with parts, labour and an outage slot. The Bluestream PdM platform is built around that path: a signal is ingested, trended, projected to the time it reaches its alarm limit, and carried through to a work order, the one stage that is still designed and not delivered. The platform is live, and as of this update it is fed by simulated signals: no plant asset is connected yet. This update sets out what is running, what is designed and what comes next, as of August 6th, 2026.
At a glance
Live · as of
- Loop stages running. Sense, stream, trend and predict run, fed by simulated signals. The fifth, act, ends in a mock adapter: the work order for Microsoft Dynamics 365 is designed, not yet delivered.
- 4 of 5
- Source types. OPC UA, Modbus TCP, standalone sensors
- 3
- Kinds of derived signal. Slope, baseline residual, efficiency ratio, hours to limit
- 4
- Alert types. Measured (at the limit) and predicted (with lead time)
- 2
- Signal source
- Simulated, no plant asset yet
- Work-order adapter
- Microsoft Dynamics 365, designed
- Projection method
- Linear, to the alarm limit, guarded
- Rule engine
- Deterministic state machine, debounce and hysteresis
- Message format
- One, versioned
- Edge buffering
- Store-and-forward
A snapshot as of August 6th, 2026. “Running” and “designed” are used strictly on this page: see What this update says, and does not say.
The loop runs end to end: on simulated signals, into a mock adapter
The platform is organised as the closed loop a predictive programme needs: sense → stream → trend → predict → act. As of this update the loop executes from one end to the other, with two qualifiers at its two ends. At the front, the signals are simulated: the gateway has the drivers, but no real asset is connected yet. At the back, the last stage stops short of Dynamics 365.
What runs at each stage
Schematic · as of Aug 6th, 2026Four stages are marked running, fed by simulated signals. The fifth executes too, but only into a mock adapter, so it is marked designed, with a dashed rule: the work order is designed for Microsoft Dynamics 365 and is not yet delivered there.
- SenseReads the sourceOne edge collector with drivers for PLCs, equipment and standalone sensors.RunningFirst real assets in progress
- StreamOne message formatEvery reading arrives in the same common, versioned format.Running
- TrendDerived signalsComputed from measured signals, and watched by the same rule engine as the raw readings.Running
- PredictHours to limitThe projected hours until the signal reaches its alarm limit. A rule on that number opens a predicted alert.Running
- ActWork orderAsset, fault mode, priority and a recommended date, for Microsoft Dynamics 365.DesignedMock adapter in development and testing
The edge collector. A single collector at the edge has drivers for three kinds of source: PLCs over OPC UA, equipment over Modbus TCP, and standalone sensors. It normalises what they send into that one message format, with store-and-forward, so readings are buffered while the network is down.
A deterministic rule engine. Incoming readings run through a deterministic state machine with debounce and hysteresis, so alerts open and clear cleanly instead of flapping.
The work order. When a rule fires, the platform is designed to raise a work order (asset, fault mode, priority and a recommended date) into Microsoft Dynamics 365, with a mock adapter used during development and testing.
The projection is one straight line to the alarm limit, with no band around it
Every measured signal can spawn derived signals: its trend slope over a window, an exponentially weighted baseline and the residual against it, efficiency ratios between two signals, and a projection of the hours remaining until an alarm limit. That last one is what the platform reports as remaining useful life: the hours until a signal reaches its alarm limit.
The same rule engine that watches raw readings watches these derived signals too, so a rule can trigger on “the projected time-to-limit has fallen below two weeks” exactly as it would on a temperature threshold.
The projection is a linear extrapolation of the trend to the limit, guarded so it only fires when the maths is sound. It is the same model shown in the interactive on the Predictive Maintenance guide in the Academy, and it is the line drawn at the top of this page.
Non-linear degradation models and confidence bands are on the roadmap, not in the platform as of this update. The planning logic stays the same either way: forecast the crossing, act before it.
A predicted alert is a different object from a measured alarm
The dashboard keeps the two apart. A measured alarm means the signal is already at the limit. A predicted alert means something is developing, and there is lead time to plan. The projection is what makes the second kind possible: it turns “vibration is rising” into “this motor crosses the alarm in about two days”. That sentence is an example of what a projection says, not a reported result.
A “why we predict this” panel shows the P–F curve and links into the Bluestream Academy guides, such as the Academy's Predictive Maintenance guide.
“The useful maintenance window is the P–F interval. A threshold alarm fires at failure; the value is catching the curve early — and turning it into a plan.”
Three horizons: now, next and later, with no dates attached
The architecture was designed with scale in mind, so the near-term work is connectivity and integrations. The back end for thousands of sensors is a Later item.
Three horizons
As published · Aug 6th, 2026Live hardware & the first pilot
- Bringing real assets online through the edge gateway, including a WAGO OPC UA connection over a secured channel
- Running the end-to-end loop on a live demo rig into Dynamics 365
- Hardening the dashboard and the operator experience
More connectivity, more integrations
- An MQTT / Sparkplug north-bound bus, so the platform can sit on a unified namespace
- Additional field protocols
- Further CMMS/ERP integrations beyond Dynamics 365
- Role-based access and multi-site organisation of assets
Scale & prescriptive analytics
- Scaling to thousands of sensors on a decoupled ingest and time-series back end
- Non-linear degradation models with uncertainty bands
- The step from predictive to prescriptive: analytics that recommend the action, not just the date
Running end to end is not the same as proven in the field
This update reports a loop that executes and a roadmap. It reports no accuracy figures, no customer site and no real asset: the loop was exercised with simulated signals, and its first horizon is still about bringing real assets online.
What this update says, and does not say
- “Running” means a stage executes and delivers where it is meant to, on simulated signals.
- Four of the five stages do. The fifth, act, executes too, but it delivers to a mock adapter and not to Dynamics 365, so it is marked designed and is not counted as running. Running does not mean a plant asset is connected or that the platform is in production at a customer site.
- The Dynamics 365 work order is designed, not delivered.
- Until it is, a mock adapter stands in for it during development and testing.
- The remaining-useful-life figure is hours to an alarm limit, not time to failure.
- It is where a straight line through the recent trend meets the limit. In general, a machine can fail before its alarm limit or run well past it.
- “About two days” and “two weeks” are examples, not results.
- The first illustrates what a projection says. The second is an example of a rule setting.
- The roadmap is an order of work, not a schedule.
- No horizon carries a date.
What runs as of this update is the arithmetic and the plumbing, on simulated signals; the first pilot is where both meet a real machine.
Since publication
Everything above describes August 6th, 2026. This note was added on September 21st, 2026 and lists what has changed since, so the snapshot can stay a snapshot:
- The WAGO controller behind the OPC UA connection named under Now was retired as a data source: the controller is end of life and its toolchain is decommissioned. In its place, a first real sensor, a low-cost vibration node, first sent readings to the platform on Aug 21st, over MQTT to a Raspberry Pi edge gateway. The rest of the demo rig is still simulated, and no customer site is connected.
- A work-order adapter for Microsoft Dynamics 365 was written and unit-tested. It has not been switched over to a live Dynamics 365 environment, so the work order is still not delivered there.
- The P10 to P90 band around the projection (P10 the earliest edge, P90 the latest; reserves reporting counts the other way) is now computed from the same fit that produces the projection. The dashboard had drawn a band since Aug 9th, but its width came from the displayed series, not from that fit. The hero figure above shows the August 6th state, without a band.
- Bluestream announced a non-binding Letter of Intent with Optec, LLC (Florida), covering field validation on real equipment. Optec's intended role is advisor and technology partner, which is why it appears in this update's tags.
- A first prior-to-posterior remaining-useful-life estimate was added behind the P10 to P90 band. It is a first version and is not yet shown to be calibrated.
- Forskningsrådet approved the underlying R&D project under SkatteFUNN, including the work on calibrated remaining-useful-life estimates that is meant to replace the straight line.
How it fits the rest of Bluestream
The PdM platform extends the Maintenance Development Toolbox: criticality and RCM decide what is worth monitoring, predictive analytics decide when to act, and the work order is designed to land where the team already works. Both use the standards and concepts taught in the Bluestream Academy.
About this update
A product update first published August 6th, 2026. It describes the platform as of that date. The page was rewritten on September 21st, 2026: the layout changed, the text now states that the signals were simulated, the lead figure counts 4 of 5 stages as running where the first version said each stage was running, the reference to Optec as technology partner moved to the dated Since publication note, and that note was added.