Time Series Analysis for Industrial Sensors

The First Question Is Always "What Does the Data Look Like?"

When something goes wrong on a production line -- an unexpected stoppage, a quality deviation, an alarm that fired at 3 AM -- the investigation always starts the same way. An engineer pulls up the sensor data and looks at it. What were the readings before the event? During? After? Did multiple sensors change at the same time, or did one lead the others?

The problem in most industrial environments is that "looking at the data" means logging into several different systems, exporting CSVs, pasting them into spreadsheets, and manually aligning timestamps. By the time you have two sensors plotted side by side, an hour has passed and the urgency of the investigation has faded.

Relay Analytics makes sensor data exploration immediate. Select your sensors, set a time range, and the system renders an interactive chart in seconds. No exports, no spreadsheets, no manual alignment. You go from question to visual answer as fast as you can think.

What Time Series Analysis Does

Time series analysis in Relay Analytics lets you plot measurements from up to 20 sensors on a single interactive chart. The chart displays sensor values over time, with each sensor rendered as a distinct trace that you can hover over, zoom into, and toggle on or off.

This is the most general-purpose analytics tool on the platform. It does not run statistical models or generate automated insights -- it shows you the raw behavior of your sensors so that you, with your domain knowledge, can see what happened and form hypotheses about why.

The typical workflow is straightforward:

1. Select the sensors relevant to your investigation.
2. Set the time range you want to examine.
3. Choose an aggregation level -- 5-second resolution for short-range detail, 15-second for medium ranges, or 60-second for longer periods.
4. Choose an aggregation function -- average for general trends, minimum to catch dips, maximum to catch spikes.
5. Visualize. The chart renders and you start exploring.

From there, you zoom in on interesting regions, compare how sensors move relative to each other, and identify the patterns that tell the story of what happened in your process.

How It Works

WebGL Rendering for Large Datasets

Industrial sensor data is dense. A single sensor at 5-second aggregation produces 720 data points per hour. Twenty sensors over an 8-hour shift is 115,200 data points. Rendering this in a standard browser charting library would be sluggish at best.

Relay Analytics uses WebGL-accelerated rendering to handle large datasets smoothly. Zooming, panning, and hovering remain responsive even when the chart contains tens of thousands of points. The system supports up to 100,000 data points per visualization, with a real-time estimation display that shows you how many points your current configuration will produce before you run the query.

Multi-Sensor Overlay

The core value of time series analysis is comparison. Plotting a single sensor over time is useful, but plotting five or ten sensors together reveals relationships that are invisible when you look at each one in isolation.

When a packaging line stops, the motor current, belt speed, and pneumatic pressure all tell part of the story. Overlaying them on the same time axis shows you the sequence of events: pneumatic pressure dropped first, then belt speed decreased, and finally the motor current spiked as the drive tried to compensate. That sequence points to an air supply problem, not a motor problem. Without the overlay, you might have investigated the motor first and lost an hour.

Z-Score Normalization

A common challenge with multi-sensor comparison is that different sensors measure different physical quantities on different scales. A temperature sensor might range from 20 to 80 degrees Celsius. A pressure sensor on the same machine might range from 2 to 6 bar. If you plot them together on the same Y-axis, the pressure trace appears as a flat line at the bottom of the chart -- completely useless for visual comparison.

Z-score normalization converts each sensor to a common dimensionless scale. After normalization, a value of 0 means "at the average," +1 means "one standard deviation above average," and -1 means "one standard deviation below." This makes the shape of each signal directly comparable, regardless of the original units.

With normalization enabled, you can see that the temperature sensor and the pressure sensor rise and fall together -- a correlation that was invisible when they were on different scales. Or you might see that they move in opposite directions, suggesting an inverse relationship worth investigating.

Aggregation Control

Not every analysis needs the same resolution. Investigating a 10-minute incident requires 5-second granularity. Reviewing a week of production data requires 60-second granularity to keep the dataset manageable.

Relay Analytics provides three aggregation levels:

You also choose how raw measurements within each aggregation window are summarized:

• Average shows the typical value -- best for general trend analysis.
• Minimum preserves the lowest reading in each window -- best for catching dips and drops that averages would smooth away.
• Maximum preserves the highest reading -- best for catching spikes and peaks that averages would hide.

This matters more than it might seem. If operators report intermittent vibration alarms but the dashboard looks normal, the issue is likely that the dashboard shows averaged values. Switching to Maximum aggregation reveals the brief spikes that triggered the alarms but were smoothed away in the average.

Real-World Example: Diagnosing a Packaging Line Stoppage

A packaging line at a consumer goods facility stops unexpectedly at 14:23 on a Wednesday. The line operator restarts it, but it stops again at 14:41. The production manager wants to know the root cause before the next shift starts.

The maintenance engineer opens Relay Analytics and selects four sensors from the packaging line: motor current, belt speed, product weight, and pneumatic pressure. She sets the time range from 14:00 to 15:00 -- thirty minutes of context before the first stoppage and fifteen minutes after the second.

At 5-second aggregation with the Average function, the chart loads in under three seconds. She immediately sees that all four signals are normal until 14:21, when pneumatic pressure begins a gradual decline. Two minutes later, belt speed drops. Motor current spikes as the drive compensates, and then the line trips.

She zooms into the 14:20 to 14:25 window by dragging across the chart. At this resolution, the sequence is unmistakable: pressure starts falling first, a full 90 seconds before the belt speed reacts. The motor current spike is a consequence, not a cause.

She switches to the Maximum aggregation function and reruns the visualization. The maximum values confirm there is no intermittent electrical spike -- the motor current increase is smooth and follows the pressure drop, not the other way around.

The root cause is in the compressed air system, not the motor or the belt. She checks the air supply manifold and finds a partially closed valve that was inadvertently bumped during morning maintenance. The total investigation takes twelve minutes from login to diagnosis. Without the multi-sensor overlay, the team might have spent hours checking electrical connections on the motor.

Key Benefits

• Immediate visual exploration. Select sensors, set a time range, and see the data. No exports, no spreadsheets, no manual timestamp alignment.

• Up to 20 sensors on one chart. Compare signals across an entire machine or process line in a single view. See relationships that are invisible when sensors are viewed in isolation.

• WebGL performance for dense data. Smooth zooming, panning, and hovering even with tens of thousands of data points. No lag, no frozen browser tabs.

• Z-score normalization for cross-unit comparison. Compare temperature, pressure, flow rate, and weight on the same visual scale. Focus on patterns and relationships, not absolute numbers.

• Flexible aggregation for any time scale. Five-second resolution for incident forensics. Sixty-second resolution for weekly trends. Three aggregation functions -- average, minimum, maximum -- to surface different aspects of the data.

• Interactive investigation tools. Drag to zoom, scroll to scale, double-click to reset. Hover to see exact values across all sensors at any timestamp. Click legend entries to show, hide, or isolate individual traces.

• Fast iteration on hypotheses. See something interesting? Change the time range, add another sensor, switch the aggregation function. Each re-render takes seconds, not minutes. The tool keeps up with the speed of your thinking.

Start Monitoring with Relay Analytics

Time series analysis is the foundation of every investigation, every optimization, and every quality review. Connect your sensors to Relay Analytics and start exploring your data visually -- from overview to root cause in minutes.