EIS: Full integration stack —
Layer 1 → 2: PhaseSeer feeds the PID controller in real time. Traditional inverter PID loops are tuned once at commissioning and react to voltage/frequency error after it happens. PhaseSeer changes this fundamentally. Because Z(ω) gives you SOC, SOH, State of Power, and State of Function continuously from the battery's own electrochemistry, the PID controller's setpoints and gain parameters are updated before a fault manifests — not in response to one. When PhaseSeer detects rising charge-transfer resistance (early degradation), it signals the inverter's PID to derate the discharge rate proactively. The PID isn't just reacting to output error anymore; it's receiving a continuous feed of battery capability from the physics layer.
Layer 2 → 3: Cyberspatial Teleseer wraps the entire OT network. The Sol-Ark inverter, BMS, TMS, and PhaseSeer measurement nodes are all addressable endpoints on a CAN/IP network. Teleseer — in its PhaseSeer-modified form — treats each of these as an identity node in a network topology. It continuously monitors traffic patterns, command sequences, and telemetry streams across the whole OT fabric. This means a spoofed CAN command telling the inverter to discharge at 3C (over the 2C limit) gets flagged as an anomaly before it reaches the inverter's PID loop, because Teleseer has learned the normal command-frequency envelope for each device.
Layer 3 → 4: Equitus IIS / ARCXA/KGNN ingests everything into a knowledge graph. ARCXA's ETL Assist connects the PhaseSeer Z(ω) stream, Teleseer's network telemetry, the BMS cell-level data, the TMS temperature readings, the inverter operating state, the grid demand signal, and the data center load demand into a single unified knowledge graph. KGNN doesn't just store these — it builds relational edges between them. A thermal spike in pack 3 of container 2, coinciding with elevated grid frequency deviation and a CAN anomaly, is a compound event that KGNN surfaces as a single node with full causal context.
Layer 4 → 5: IBM watsonx executes NNX-portable decisions at the edge. The NNX-exported models trained on EIS data run identically on IBM watsonx (enterprise governance, FedRAMP, audit logging) and on the edge compute node inside the container. The decision — reroute load to pack 5, derate inverter 3 to 70%, alert the Equitus architect — is made in milliseconds at the edge, then logged and audited on watsonx. The AI doesn't phone home to the cloud to decide; it decides locally and reports up.
Human-in-the-loop (HITL) Equitus architect sits at the top of the stack. The system escalates to a human only when compound events exceed its confidence threshold — a combination of anomalies it hasn't seen before, or a decision that requires cross-site authorization. Everything else runs autonomously.
Full architecture diagram, demonstrates the data flow as an interactive control loop. Every node in that diagram is clickable and will drill into the mechanism behind it. Now the dynamic control loop — this is the part that's hardest to show in a static diagram, because the critical insight is the feedback direction: PhaseSeer doesn't just report upward to KGNN, it simultaneously feeds back down to the PID controller to modify inverter behavior in real time. The animated flows show the two critical feedback paths that make this architecture different from conventional BESS control:
The PhaseSeer → PID feedback path (green dashed, flowing back down to the inverter) is the architectural innovation. Most inverter PID loops only have one feedback source — the output voltage/frequency measurement. This system has a second, deeper feedback source: the electrochemical state of the battery itself. PhaseSeer continuously re-evaluates what the battery is capable of delivering and adjusts the PID's setpoints accordingly. If SOH drops on pack 3, the PID controller for inverter 3 gets a lower discharge ceiling before the voltage actually sags. Prevention rather than reaction.
The watsonx → inverter action path (teal dashed, looping from the bottom back up to the inverter) is the AI decision closure. NNX models running on IBM watsonx evaluate the compound event from KGNN — the full picture of Z(ω) state, Teleseer anomaly score, grid frequency, thermal readings, and BMS alerts — and issue an action: reroute load to another pack, derate this inverter to 70%, isolate the anomalous CAN node, or escalate to the Equitus architect. That action travels back down to the physical layer as a new PID setpoint, closing the loop.
The top diagram shows the physical layer integrated with the AI overlay — the dashed purple telemetry lines are the key innovation. Every BESS pack, the inverter, TMS, and grid connection continuously broadcast to PhaseSeer. The physical and cyber layers aren't separate systems bolted together — they share a single identity layer where each battery cell is a cyberspace-addressable sensor.
The bottom diagram breaks out the control plane in full. The signal flow runs top to bottom:
Physical sensors → PhaseSeer computes Z(ω) in real time → two parallel paths emerge: NNX model inference (which runs identically at the edge or on IBM watsonx) and ARCXA/KGNN ingesting all the other SLED data — SCADA, fleet telematics, grid dispatch schedules — alongside the impedance stream. Both paths converge into the living knowledge graph, which drives three distinct output functions: energy dispatch (peak shaving, frequency response), fault detection with auto-rerouting, and cybersecurity anomaly detection via the Teleseer OT/IT monitoring layer.
The EIS outcome layer at the bottom is what the DoD/SLED customer actually buys: uptime, mission continuity, compliance reporting, threat response, and grid resilience — all from one integrated platform. Every clickable node will drill into the underlying mechanics if you want to explore any layer further.
No comments:
Post a Comment