Analytics & ML · Study deck
Edge Acquisition: Device Architecture
An accelerometer stream can overwhelm a small gateway before any model runs.
Data Dora is your guide for this deck.
After studying this chapter
Learning objectives
Choose acquisition hardware from the source, timing, storage, and connectivity requirements.
- Device categories depend on computing resources and connectivity.The factory combines Non-IP temperature sensors, Small IP cameras, and a Big Thing SCADA server.
- A gateway gives Non-IP sensors a route to internet services.Modbus RS-485 temperature sensors need translation before their measurements can reach cloud infrastructure.
- Acquisition must preserve the meaning of each measurement.Each bearing reading must retain source identity, event time, units, quality, and device state.
- Local processing has limits as well as benefits.Delay, memory, power, and maintenance determine which decisions can remain near the machine.
Major section
Start With the Story
The bearing example separates urgent local action from analysis using remote history.
- The controller handles urgent action near the bearing.The remote service can compare the change with months of history while the local controller can stop the machine.
- A broken link must leave a defined local response.Waiting for every raw value to reach the cloud could turn a useful warning into late evidence.
- A decision record connects the input to the result.The matched bearing runs must retain code version, rule, input, time, and action for later review.
- Failure tests expose the limits of local action.Dropped samples, full storage, and competing workloads test whether uncertainty remains visible when the remote path disappears.
Major section
From Vibration to Digital Codes
Follow the vibration example from acceleration to tiny mechanical movement and a digital reading.
- The baseline displacement is only 31.1 pm.The 1 pF differential sense capacitor has a 31.1 aF change at 0.05g and the stated 20 kHz resonance.
- The alert produces ten times the physical change.The 0.5g reading corresponds to 311 pm of motion and 311 aF of capacitance change.
- The digital output has a finite measurement step.The stated ±4g, 16-bit output changes by 122 µg/LSB, linking the physical signal to discrete codes.
- The two vibration levels produce distinguishable digital readings.The baseline is near code 410, while the alert is near code 4,096 in the chapter’s example.
Major section
Edge Data Acquisition Basics
Local acquisition reduces what travels upstream while preserving information needed for useful decisions.
- The edge is the place where sensing happens.Factory instruments and roadside cameras can collect observations away from the central servers.
- Local processing selects the information worth forwarding.The security camera detects motion and uploads relevant clips instead of sending its entire raw stream.
- Selective forwarding changes the upstream traffic requirement.The camera can generate 1GB of video per hour while forwarding only the 5-second clips chosen by motion detection.
- The processing split still needs a resource budget.Local power, memory, context, and upkeep limit how much analysis the device can own.
Major section
Key Concepts
Sampling, buffering, and forwarding have distinct jobs inside an acquisition device.
- Sensor buses carry measurements into the acquisition device.I2C, SPI, UART, and ADC interfaces connect sources to the microcontroller or gateway.
- A ring buffer can bound the recent history in memory.Its fixed-size circular structure can retain a rolling window of readings without dynamic memory allocation.
- Timer interrupts keep sampling intervals consistent.The microcontroller can trigger sensor reads regularly without spending its time on busy-wait polling.
- DMA can move peripheral data without processor intervention.ADC or sensor-bus transfers can reach memory while the processor remains available for other tasks.
Major section
Three Categories of Things
This diagram compares connectivity paths; trace Big Things, Small IP Things, then the gateway route for Non-IP Things.
- Big Things combine compute capacity with a direct IP path.The first route covers servers and industrial PLCs that can connect without a translating sensor gateway.
- Small IP Things retain their own internet connectivity.The camera route uses Wi-Fi or cellular while operating with tighter local resources than the server.
- Non-IP Things depend on a translating gateway.Simple sensors end at a field bus or short-range link before their data reaches cloud infrastructure.
- The connection route defines the category better than physical size.The factory’s Modbus sensors, Wi-Fi cameras, and SCADA server follow three different acquisition paths.
Activity 1 · Draw it
✎ Connect the factory sources
I want you to draw the internet path before you assign a device category.
Sketch the chapter factory with Modbus RS-485 temperature sensors, Wi-Fi IP cameras, and a central SCADA server. Label each device category and show which path requires a gateway.
3 minutes · Pen and paper · Answer: Activity 1
Deck summary
Key takeaways
A useful acquisition architecture preserves meaning and bounds the local decision.
- Device capability determines which acquisition jobs can stay local.The factory’s sensors, cameras, and server differ in both computing resources and routes to internet services.
- Gateway translation must preserve the measurement’s meaning.The source identity, event time, units, and quality remain necessary after data crosses a protocol boundary.
- Local processing reduces dependence on remote response.The bearing controller can handle urgent action while a remote service reviews longer operating history.
- Recovery evidence belongs in the architecture review.Matched runs check local action and later record recovery after dropped data, full storage, or an unavailable remote path.
Retrieval practice
Recall check
Data Dora says: answer from memory, then check your reasoning.
Q1A factory has Modbus RS-485 temperature sensors, Wi-Fi IP cameras, and a central SCADA server. How does the chapter's device taxonomy classify them?
Show answer
Answer: A The taxonomy sorts devices by connectivity and compute: Non-IP Things, Small IP Things, and Big Things.
Print reference
Answers
Answer key.
- A · The taxonomy sorts devices by connectivity and compute: Non-IP Things, Small IP Things, and Big Things.
Print reference
Activity 1 answer
Model answer.
Draw it: The temperature sensors are Non-IP Things and require gateway translation. The Wi-Fi cameras are Small IP Things with an IP path. The SCADA server is a Big Thing. The sketch distinguishes the translated sensor route from the IP-connected devices.