Automation, sensors and predictive models

From monitoring to control

A growing chamber displays temperature continuously. This is monitoring. A forecast estimates its temperature later. A controller commands a fan when a chosen condition is met. These functions use related information, but they do different jobs.

Automation connects information to actions. In a conceptual feedback system, a sensor supplies an indication, a decision rule produces a command, an actuator changes the process, and subsequent measurements help assess the result. The diagram shows that loop. Sending a command is not proof that the equipment responded or that the desired production condition was achieved.

Measure the intended condition

A temperature reading describes a defined location, time and method. It may not represent every place in a chamber, or the temperature within a plant or food product. Sensor placement, response and the relationship between indication and reference values matter for the interpretation.

Calibration establishes a relationship under specified conditions, with measurement uncertainty. A comparison appropriate for one temperature range or setup need not establish adequacy in another. NIST distinguishes a documented calibration chain from fitness for a particular decision. A useful system names the quantity it needs, rather than assuming the displayed number describes the whole process.

A rule and an uncertain reading

Imagine a teaching controller with the invented rule: command ventilation when the indicated temperature exceeds 25°C. This setting is stipulated to explain a decision rule, not recommended for a crop or used to establish food safety. If the indication is 24.6°C, the rule does not trigger on that indication.

Now suppose the measurement is reported with a stated interval of 24.6 ± 0.8°C for the relevant temperature. Its endpoints are 23.8 and 25.4°C. That interval crosses the rule’s threshold. It does not establish that the relevant temperature is definitely below 25°C. No confidence probability or complete uncertainty budget has been supplied; the example shows why the decision’s evidence needs attention.

Prediction adds assumptions

A predictive model can estimate a future condition using current information and assumed future inputs. A controller might use such an estimate to act before a limit is reached. The estimate remains a prediction, and its relevant errors need evaluation against observations not used for model fitting.

Changes in weather, loading, sensor quality or equipment response can undermine a previous relationship. A model should not silently substitute its output for a missing measurement and present it as observed. Keep readings, estimates, commands and verified outcomes identifiable so faults and disagreements can be investigated.

Count the defined resource

For another invented case, assume a fan draws a constant 0.20 kW whenever it operates. Three operating hours consume 0.20 × 3 = 0.60 kWh. Continuous operation for six hours would consume 1.20 kWh. The difference is 0.60 kWh of fan electricity under these assumptions.

That calculation does not establish equal growing conditions or a reduction in total production energy. Heating, cooling, other devices and crop outcomes are outside this boundary. A useful comparison measures the complete outcome of interest and makes the included loads and time period explicit. A smaller device electricity total is only one component of a production-system assessment.

Verification and failure response

A failed sensor, stale reading, disconnected communication link or stuck actuator can interrupt the loop. The system needs a defined response and appropriate human support for such conditions. Maintenance must address equipment-specific hazards and competence; a software stop is not proof of safe physical isolation.

FAO’s 2022 precision-agriculture case study analysis distinguished observing, measuring and responding, and described evidence and infrastructure limits. Automation should therefore be evaluated by useful decisions and verified production outcomes, alongside costs, energy, maintenance and skills. Detailed growing conditions and equipment design belong in their relevant production and engineering articles; the common control loop connects them.

Worked examples

Check your understanding

1. A controller logged a fan-start command. Has actual airflow been established?

   Show answer and reasoning

   No.

   A command and a successful physical response are different records; the actuator or airflow may need verification.

2. A model fills a gap left by a failed sensor. How should that output be described?

   Show answer and reasoning

   As an estimate, with its assumptions and uncertainty.

   A prediction must not be labelled as a direct observation.

3. Why can reduced fan electricity coexist with unchanged or higher total energy use?

   Show answer and reasoning

   Other system loads may change.

   The device-only account excludes heating, cooling and other loads; total-system effects need their own assessment.

Key ideas

• Keep observations, predictions, commands and verified responses distinct.
• Define measurement conditions and uncertainty around decisions.
• Evaluate complete production outcomes and failure support, with explicit resource boundaries.