SoftTech AI Registry
Imubit · Imubit Platform

Autonomous Industrial Process Setpoint Control — DLPC closed-loop setpoint execution

In Closed-Loop Mode, the Imubit Platform uses a control policy learned through reinforcement learning, captured in a deterministic controller, to determine optimising industrial process-control setpoints from plant/process data and write them autonomously into the plant's existing DCS/APC control environment, within predefined constraints and without human confirmation for each action. Operators retain supervision and control authority. Open-Loop (advisory) Mode is excluded. The public evidence does not establish direct valve, actuator, motor or other equipment commands, a watchdog/MV-interface validation step, or the downstream PID relationship.

Recorded characteristics

Function
Imubit states the platform "takes them all the way to autonomous execution through learning process models" and is "powered by Deep Learning Process Control® (DLPC), our patented approach to nonlinear industrial process modeling, control, and optimization." Phase 3 — Control: "Decisions executed autonomously, gaps closed. Validated models extend into closed-loop execution without changing tools or architecture. Reinforcement learning across all process regimes is captured in a deterministic controller your engineers can certify and your operators trust — steering decisions to maximize objectives within explicit constraints, with operators holding full authority." An Imubit article states plants can allow "the model to write setpoints directly to the DCS once trust is established." Approved decision chain (stages kept separate): plant/process data → control policy derived through reinforcement learning, captured in a deterministic controller → optimising process-control setpoint → autonomous write to the existing DCS/APC environment → live industrial-process control. The deployed controller embodies a policy learned through reinforcement learning; the evidence does not establish that the production controller continuously learns during operation (Imubit states only that "models retrain as your plant changes"). Controlled object: a DLPC-generated industrial process-control setpoint passed into the plant's existing DCS/APC control environment. Closed-Loop Mode only.
Data access
Nonlinear dynamic models built from the plant's historical data, guided by first principles; native historian and API integration; on-premise deployment within plant networks. Exact live data inputs per deployment are not publicly established.
Actions
Can take actions
External actions
Yes
Human confirmation
Not required
Permission basis
Not established
Administrative control
Imubit lists: "Explicit constraint enforcement", "Operator supervision and control authority", "Scoped automation across selected handles or units", "Version control and auditability", "Model and data governance". Engineers define operational decisions, measured outcomes, constraints and limits, and strategic objectives. Deployment Mode (Open-Loop or Closed-Loop) is specified per site; the reseller agreement requires "appropriate safeguards, operator training, and change management processes" before Closed-Loop Mode is enabled at a customer site. Safety boundary: the AI decision determines the optimising setpoint within constraints. No plant safety authority is attributed to Imubit's AI: safety instrumented systems (SIS), emergency shutdown, plant interlocks, DCS protective logic, PID safeguards and regulatory safety systems are explicitly not attributed.
Default state
Conditional
Availability
Imubit reports "100+ closed-loop AI applications deployed worldwide". On-premise deployment within plant networks; DLPC On Prem "Enable[s] supervised, constraint-aware automation when activated". Industries named: process industries; cement and building materials; chemicals, petrochemicals and polymers; mining, minerals and metals; oil and gas.
Licensing
Sold directly and through resellers; deployment mode (Open-Loop or Closed-Loop) is specified per site in the customer agreement, and resellers need additional certification to offer Closed-Loop Mode. Pricing is not publicly established.
External model or provider
Imubit's own Deep Learning Process Control® (DLPC) technology. No third-party model provider is established.
Limitations and uncertainty
Not publicly established: (1) exact RL algorithm; (2) exact neural architecture; (3) per-setpoint confidence; (4) per-setpoint explanation; (5) exact reward/objective weighting; (6) universal fail-safe state; (7) watchdog-trip behaviour; (8) network-loss behaviour; (9) stale-data behaviour; (10) sensor-failure behaviour; (11) automatic rollback; (12) exact manual-override mechanism; (13) universal setpoint movement limits; (14) universal control interval; (15) maximum MV count; (16) maximum plant scope; (17) complete audit retention; (18) exact authentication/permission mechanism; (19) DCS watchdog / manipulated-variable (MV) interface validation of setpoints before execution; (20) downstream PID relationship; (21) the detailed physical execution chain below the DCS/APC environment, including any direct valve, actuator, motor or other equipment command; (22) whether the production controller learns during operation (only model retraining is stated). Default state: conditional — Deployment Mode is specified per site and DLPC On Prem automation operates "when activated"; no universal default is established. Product identity: the current product page names the "Imubit Platform"; the same page was previously titled "Operations Studio". Excluded: Open-Loop Mode (AI-generated recommendations that operators decide whether and how to implement); advisory mode; forecasting and prediction alone; simulation and what-if analysis; dashboards and performance monitoring; customer-authored deterministic APC/DCS logic; the customer's DCS, APC, PID and safety systems themselves. Monitoring: automated monitoring deliberately withheld. Imubit's Terms of Use restrict "any scraper, crawler, spider, robot or other automated means of any kind to access or copy data on the Platform", which may apply. A two-fetch test of the product page on 3 Oct 2026 succeeded (200, no redirect, 5,045 characters, matching fingerprints); this establishes technical monitorability only and does not override that restriction.

Evidence