← Back to Field Notes

April 8, 2025 · The Bloomfield Team

How to Collect Machine Data Without Spending Six Figures

The machine monitoring market wants you to believe that collecting data from your shop floor requires a $150,000 platform, a six-month implementation, and an annual subscription that costs more than your highest-paid machinist. For large operations running 50 machines across three shifts, that kind of investment can make sense. For a 15-machine job shop doing $6 million in annual revenue, the math falls apart before the demo is over.

The good news: 80% of what most small manufacturers need from machine data can be captured with equipment costing under $3,000 and a setup process measured in days rather than months.

What You Actually Need to Know

Before buying any hardware, clarify what questions machine data should answer. Most shops need three things from their machines: Is the machine running right now? How long has it been running versus idle today? How does the actual cycle time for this part compare to the standard cycle time in the ERP?

Those three data points, collected consistently across every machine on the floor, will change how a shop schedules work, quotes jobs, and identifies capacity that is currently hidden. Everything beyond that, spindle load monitoring, vibration analysis, predictive maintenance algorithms, is valuable. It is also secondary. Get the foundation first.

Three Tiers of Machine Data Collection

Tier 1: Current sensing with off-the-shelf hardware ($200 to $500 per machine). A split-core current transformer clamped around the power feed to a CNC spindle motor can tell you whether the machine is running, idle, or off. This is the same technology electricians use to measure amperage, packaged into an IoT device with a wireless transmitter. Companies like Shingle and MachineMetrics offer entry-level sensors in this category, and several open-source hardware projects exist for shops comfortable with basic electrical work. The sensor detects current draw. When the spindle is cutting, amperage is high. When the machine is idle, amperage is low. When the machine is off, amperage is zero. That binary signal, running versus not running, collected every 30 seconds, produces utilization data that is accurate to within 2% of actual spindle runtime.

Tier 2: MTConnect or OPC-UA from the controller ($0 to $1,000 per machine). Most CNC machines built after 2010 support MTConnect, an open protocol that streams data directly from the machine controller. Haas machines have it built in and it can be activated in the settings menu. Fanuc controllers support it through a FOCAS adapter. Mazak, Okuma, and DMG Mori all offer MTConnect options, though some require a hardware adapter that costs $500 to $1,000. The data from MTConnect is richer than current sensing. It includes spindle speed, feed rate, program name, tool number, alarm status, and part count. This level of detail supports real scheduling decisions and accurate cycle time tracking.

Tier 3: Edge computing with a Raspberry Pi or industrial PC ($500 to $2,000 per cell). An edge computer sits on the shop floor, collects data from multiple machines via MTConnect or current sensors, stores it locally, and pushes summaries to a dashboard or database at configurable intervals. A Raspberry Pi 4 running a lightweight data collection stack (Node-RED, InfluxDB, Grafana) can handle four to six machines and cost under $200 for the hardware. For environments with more dust, vibration, or temperature variation, an industrial-rated edge PC from OnLogic or Advantech runs $800 to $1,500 and handles the same workload with better reliability.

The Data Architecture That Scales

Regardless of which tier a shop starts with, the data needs to land in a structured format that can grow. A simple time-series database, InfluxDB is free and open-source, stores timestamped records of machine state and cycle data. A visualization layer on top, Grafana is also free, turns those records into dashboards that the production team can actually read.

The total cost for a 10-machine shop using Tier 1 sensors, a single Raspberry Pi as the data collector, and InfluxDB plus Grafana for storage and visualization: approximately $2,500 to $3,500 in hardware and one to two weeks of setup time for someone comfortable with basic networking and Linux commands.

For a broader perspective on connecting shop floor systems, see our guide to ERP and AI integration.

What to Do With the Data

The first month of machine data reveals patterns that most shop owners suspect but have never quantified. Machine A runs at 72% utilization. Machine B runs at 41%. The gap between those two numbers contains the capacity to take on three more jobs per week without adding a machine or an operator.

Cycle time data shows which parts run faster than standard and which run slower. Over 90 days, a shop accumulates enough actual cycle time data to rebuild its estimating standards on real performance rather than original guesses. That data feeds directly into more accurate quoting, which feeds directly into better margins.

Downtime categorization, even at the simple level of "running versus not running," shows where the dead time lives. First shift startup is slow because operators spend 25 minutes finding tooling. Mid-shift gaps happen because material staging does not align with the schedule. End-of-shift slowdowns start 45 minutes before the bell because jobs get held for the next day rather than started and interrupted.

None of these patterns are visible without data. All of them are fixable once they are visible.

Start With One Machine

The fastest way to prove the value of machine data in your operation is to instrument one machine for 30 days. Pick the machine with the most contested utilization claim. Is it really running 80% of the time like the operator says, or is it closer to 55% like the scheduler suspects? Thirty days of data will answer that question definitively, and the answer will fund the rest of the rollout.

For shops ready to build on machine data with AI-powered analysis and decision support, the data collection layer is the prerequisite. Start collecting. The intelligence comes after.