AI Assistant for Industrial IoT Analytics

Your Data Knows the Answer -- You Just Need a Way to Ask

Industrial facilities generate enormous volumes of sensor data. Temperature, pressure, vibration, flow rate, weight -- every measurement is recorded, timestamped, and stored. The information needed to diagnose a problem, optimize a process, or prevent a failure is almost always already in the data.

The bottleneck is not data collection. It is data access. Answering a question like "What caused the vibration spike at 2 PM?" typically requires a data engineer to write SQL queries, a process engineer to interpret the results, and several rounds of back-and-forth to refine the analysis. Simple questions take hours. Complex investigations take days.

Relay Analytics removes that bottleneck with a conversational AI assistant built specifically for industrial sensor data. You ask questions in plain language. The AI selects the right analytical tools, runs the calculations, and returns answers with charts, tables, and explanations -- all in a single conversation.

What the AI Assistant Does

The AI Assistant is an agent with specialized analytical tools, not a generic chatbot. It has direct access to your sensor data and can perform the same analyses that would normally require manual configuration across multiple screens.

Correlation analysis -- Ask "Which sensors are correlated?" and the assistant calculates Pearson correlation coefficients between every pair of selected sensors. It returns a correlation matrix and interprets the results: which pairs are strongly related, which are independent, and which show unexpected relationships.

Clustering -- Ask "Group these sensors by behavior" and the assistant runs hierarchical clustering to identify which sensors behave similarly. It reports the groups and explains what the grouping implies about your process.

Rolling correlation -- Ask "When did the relationship between sensor A and sensor B change?" and the assistant tracks correlation over time around a specific reference point. It identifies the moment a relationship shifted, broke down, or reversed.

Historical data queries -- Ask "What was the average temperature last week compared to this week?" and the assistant writes and executes SQL against your full historical dataset. It handles aggregations, comparisons, percentile calculations, time bucketing, and ranking -- all from a natural language request.

Chart generation -- Ask "Show me a line chart of these sensors" or "Create a correlation heatmap" and the assistant generates the visualization inline in the conversation. Line charts, bar charts, scatter plots, and heatmaps are all available.

CSV export -- Ask "Export this data to CSV" and the assistant generates a downloadable file attached directly to the conversation.

Raw data inspection -- Ask "Show me the actual readings from sensor X between 10:00 and 10:30" and the assistant retrieves the specific measurement values for spot-checking.

Every response streams in real time -- you see the assistant's reasoning as it works, not just a final answer after a long wait.

How It Works

When you start a conversation, you select the sensors and time range you want to analyze. The assistant uses this as its working context -- it knows which devices to query and what time period to focus on.

As you ask questions, the assistant selects the appropriate tool for the job. A question about relationships triggers correlation analysis. A question about historical trends triggers a SQL query. A request for a chart triggers visualization generation. You do not need to know which tool to use -- the assistant maps your intent to the right capability.

Domain Knowledge from Your Documents

The assistant becomes significantly more useful when it has access to your technical documentation. Relay Analytics includes a Knowledge Base where you upload equipment manuals, specification sheets, maintenance procedures, and fault diagnostic guides.

When you ask a question, the assistant searches your uploaded documents for relevant passages and combines that domain knowledge with live sensor data analysis. Instead of generic statistical observations, it provides answers grounded in your specific equipment and processes.

For example, without documentation, the assistant might report: "Sensor A and Sensor B have a correlation of -0.78." With an uploaded equipment manual, it can say: "The oil temperature (Sensor A) and coolant flow rate (Sensor B) have a strong inverse correlation of -0.78. According to the KX-7700 operation manual, the expected relationship is Oil Temp = 112.7 - (1.33 x Coolant Flow). The observed correlation is consistent with normal heat exchanger operation."

The system supports PDF, text, and Markdown files up to 10 MB. Documents are automatically processed and indexed for semantic search -- the assistant finds relevant passages based on meaning, not just keyword matching.

Contextual Understanding

You can provide the assistant with a reusable context block that describes your machine, sensor layout, normal operating conditions, and analysis goals. This context is sent with every message, giving the assistant a persistent understanding of what it is analyzing and why.

A well-written context transforms the assistant from a statistical calculator into a domain-aware analyst. It knows that sensor WH-001 is a weighing hopper, that CW-001 is a checkweigher downstream, and that a correlation between them is expected because they measure the same product at different stages.

Real-World Example: Diagnosing a Vibration Spike

A maintenance engineer at a chemical processing plant notices that a vibration alarm fired on Compressor 7 at 14:12 on Thursday. The alarm cleared on its own after eight minutes. The compressor is still running, but the engineer wants to understand what happened and whether it will recur.

She opens the AI Assistant, selects six sensors from Compressor 7 -- vibration, oil temperature, oil pressure, coolant flow, discharge pressure, and motor current -- and sets the time range to 13:00-16:00 to capture the event with ample context.

She has previously uploaded the compressor's operation manual and the cooling system specification sheet to the Knowledge Base.

Her first question: "What caused the vibration spike at 2 PM?"

The assistant runs correlation analysis across all six sensors and identifies that vibration has a strong inverse correlation (-0.82) with coolant flow rate during the analysis period. It then runs rolling correlation around 14:12 and finds that the correlation weakened starting at 14:08 -- four minutes before the alarm.

Next, the assistant searches the uploaded equipment manual and finds a section on vibration troubleshooting that lists "reduced coolant flow" as a common cause of bearing vibration in this compressor model. The manual specifies that coolant flow below 12 L/min can cause thermal expansion of the bearing housing, which increases vibration.

The assistant checks the raw data and confirms that coolant flow dropped from 15.3 L/min to 11.1 L/min at 14:08 and recovered to 14.8 L/min at 14:20 -- aligning with the vibration spike timeline.

It responds with a summary: the vibration spike was caused by a temporary reduction in coolant flow that allowed the motor bearing housing to thermally expand. The coolant flow dip lasted 12 minutes and the vibration followed with a 4-minute lag, consistent with the thermal time constant described in the equipment specification. The bearing is likely undamaged since the vibration returned to baseline, but the engineer should investigate why coolant flow dropped -- possible causes include a partially blocked filter or a momentary control valve fault.

The entire investigation -- from question to root cause to recommended action -- takes five minutes.

Key Benefits

• Natural language access to complex analysis. Ask questions the way you think about them. "Which sensors spiked last Tuesday?" or "Compare this week to last week" -- no SQL knowledge or statistical expertise required.

• Multiple analytical tools in one conversation. Correlation, clustering, rolling correlation, historical queries, charting, and data export are all available without switching screens or tools.

• Domain-specific answers from your documentation. Upload equipment manuals, spec sheets, and maintenance procedures. The assistant references them automatically, citing specific thresholds, formulas, and procedures from your documents.

• Real-time streaming responses. See the assistant's reasoning as it works. No waiting for a spinner to finish -- the response builds in front of you, token by token.

• Charts and exports inline. Visualizations appear directly in the conversation. CSV files are attached for download. No context switching to separate dashboards or export tools.

• Historical depth with SQL access. The assistant queries your full measurement history, not just recent data. Compare periods, calculate percentiles, rank sensors by variance, and detect long-term trends across weeks or months.

• Contextual continuity within a session. The assistant remembers what you discussed earlier in the conversation. Ask a follow-up question and it builds on previous results rather than starting from scratch.

• Accessible to non-technical users. Operators, shift supervisors, and quality engineers can run sophisticated analyses without waiting for data team availability. The assistant handles the technical complexity behind the scenes.

Start Monitoring with Relay Analytics

The AI Assistant is available to every Relay Analytics user. Connect your sensors, upload your equipment documentation, and start asking questions about your data. From simple lookups to complex multi-sensor diagnostics, the assistant meets you at your level of expertise and delivers answers grounded in your actual measurements.