Long-wave infrared (LWIR) thermography has become increasingly useful in industrial automation because it can measure surface temperature without physical contact. Uncooled microbolometer focal plane arrays (FPAs) offer low power consumption, compact construction and continuous operation, making them suitable for machine monitoring, process inspection and condition-based maintenance.
However, designing a compact thermal vision system is not simply a matter of selecting the highest-resolution sensor available. Engineers must balance three closely connected factors: spatial resolution, thermal sensitivity and optical cost. Increasing performance in one area can impose constraints on another, particularly when the camera must fit inside a small machine enclosure or robotic system.
Why Compact Thermography Is an Engineering Trade-Off
An industrial thermal camera is often treated as a temperature sensor rather than a conventional visual camera. Its primary function is to convert infrared radiation from a target surface into calibrated temperature information that can be evaluated by a PLC, industrial PC, vision controller or other automation system.
This distinction is important when specifying a camera. A high-resolution thermal image is of limited value if the system cannot distinguish the temperature difference that matters to the process. Likewise, extremely low NETD does not compensate for insufficient spatial resolution when the target is small.
In my view, the most practical approach is therefore to define the smallest thermal feature that must be detected first. Once the required field of view, measurement distance, minimum target size and temperature difference are established, sensor resolution and optics can be selected around those requirements rather than the other way around.
Pixel Pitch and Spatial Resolution
The physical dimensions of the microbolometer array have a direct influence on the optical design. Pixel pitch determines the angular sampling of the detector when combined with the selected lens and working distance.
Reducing pixel pitch allows more detector elements to be placed within a given sensor format, or allows a smaller active detector area to provide a comparable number of pixels. This can help reduce the required image circle and, in some applications, permit a more compact optical assembly.
A 640 x 480 detector with a 12 um pixel pitch, for example, has a smaller active area than a detector with the same pixel count but a larger pitch. This can be advantageous when the camera must be installed inside a restricted enclosure.
The important point, however, is that smaller pixels do not automatically create better system-level resolution. The lens, detector response, modulation transfer function, working distance and target size all contribute to the final image quality. Pixel pitch should therefore be considered as one component of the optical system rather than an independent performance indicator.
The Optical Cost of Higher Resolution
LWIR optics are subject to different material and transmission requirements than visible-light lenses. Germanium and other infrared-transmitting materials can add significant weight and cost, particularly when large apertures and complex lens configurations are required.
A smaller detector format can reduce the required image circle, which provides an opportunity to reduce lens size. For stationary machine-vision installations, this can be particularly valuable because the camera may have to fit behind protective windows, inside electrical cabinets or between mechanical assemblies.
There is also a practical mechanical benefit. A smaller optical assembly can reduce the mass and mounting requirements of the camera. This becomes more important when thermal cameras are installed on moving robots, inspection mechanisms or compact OEM equipment.
The engineering target should therefore not be "maximum resolution at any cost." It should be sufficient resolution with the smallest optical system that meets the measurement requirement.
Thermal Sensitivity Is More Than a NETD Number
Spatial resolution determines whether a thermal feature can be separated geometrically, while thermal sensitivity determines whether its temperature difference can be detected.
Noise Equivalent Temperature Difference (NETD) is commonly used to describe the thermal sensitivity of an infrared detector. Lower NETD values indicate a smaller temperature difference that can theoretically be distinguished under specified measurement conditions.
For industrial inspection, NETD in the 20 mK to 30 mK range can be useful when the application involves relatively small temperature gradients. Examples include monitoring thermal behavior around casting processes, detecting localized heating on moving components or identifying developing differences across a manufactured surface.
Nevertheless, NETD should not be interpreted as guaranteed measurement accuracy. Emissivity, reflected radiation, atmospheric transmission, detector temperature, lens transmission and calibration all affect the final temperature measurement.
This is one area where I would recommend avoiding a specification-sheet-only approach. A camera with excellent laboratory NETD may deliver less useful process information than a camera with slightly higher NETD but better optical matching, calibration and integration with the production system.
LWIR Operation and Environmental Compensation
The 8 um to 14 um atmospheric transmission region is widely used for industrial LWIR imaging because it provides a practical measurement window for many applications. Atmospheric absorption still depends on distance, humidity and environmental conditions, so the optical path cannot be ignored when specifying a thermal measurement system.
Uncooled microbolometers also respond to changes in their own operating temperature. Without compensation, detector drift can affect radiometric stability as the camera environment changes.
Modern systems therefore use calibration and software-based compensation techniques to correct for detector response changes. For industrial automation, this function is particularly relevant when cameras operate continuously across changing ambient conditions.
It is also important to distinguish thermal sensitivity from absolute temperature accuracy. A low NETD indicates sensitivity to temperature differences, while absolute accuracy depends on calibration and measurement conditions. These two specifications should not be treated as interchangeable.
Radiometric Data Must Reach the Control System
A thermal camera installed in an automated process is only useful when its measurement data can reach the control or analysis layer with predictable timing.
Streaming full-frame radiometric data at high frame rates can create substantial data throughput requirements. The system architecture must therefore consider not only the detector but also the interface, processor, storage or host computer and software pipeline.
For stationary industrial installations, Gigabit Ethernet is a practical architecture because it supports relatively long cable runs and integration with industrial computing infrastructure. When Power over Ethernet (PoE) is supported, power and communications can also be combined into a single cable, simplifying camera installation.
Hardware triggering is equally important for applications where thermal measurements must correspond to a specific machine position, production cycle or inspection event. In these systems, deterministic triggering can be more valuable than simply increasing frame rate.
GigE and PoE for Factory Integration
GigE Vision-based architectures are well suited to fixed industrial installations where the camera is separated from the processing computer.
The camera can transmit thermal image data over Ethernet while the host system performs image analysis, temperature calculations, alarming and data logging. This architecture also allows the processing hardware to be located away from heat, vibration or other environmental constraints around the machine.
PoE can further reduce wiring complexity by carrying power and data through the same Ethernet connection. For production equipment containing multiple sensors and cameras, reducing the number of individual cables can simplify installation and maintenance.
The limitation is that Ethernet hardware introduces additional physical and processing requirements compared with a direct sensor-to-processor connection. For extremely space-constrained embedded designs, another interface may be more appropriate.
MIPI for Embedded and OEM Designs
MIPI-based interfaces are increasingly relevant when the thermal sensor is integrated directly into a compact embedded vision platform.
Instead of routing data through a conventional Ethernet camera architecture, MIPI can provide a direct high-speed connection between the imaging device and a host processor or image signal processor. This reduces the need for Ethernet transceivers, connectors and associated hardware.
The resulting architecture can be attractive for compact OEM machines, autonomous equipment and UAV payloads where board area, power consumption and physical mass are tightly constrained.
The trade-off is integration complexity. MIPI-based systems generally require more attention to hardware design, drivers, synchronization and software integration. For an OEM developer with control over the complete electronics platform, that complexity may be justified. For a conventional factory installation, Ethernet may still provide a simpler overall system architecture.
Matching the Camera to the Machine
The correct thermal camera should be selected from the process requirement outward.
A useful engineering specification should include at least the target dimensions, working distance, field of view, minimum temperature difference, expected temperature range, required measurement accuracy, frame rate, triggering requirements and available installation space.
For example, detecting a large hot bearing housing does not require the same optical configuration as measuring a narrow thermal band on a moving component. Similarly, monitoring whether a process exceeds a temperature threshold is fundamentally different from performing detailed radiometric analysis across a small surface.
This is why camera resolution alone is a poor basis for equipment selection.
Compact Does Not Mean Compromised
The development of smaller-pitch microbolometers, compact LWIR optics and high-speed digital interfaces has changed the practical size of industrial thermal imaging systems.
A 12 um detector architecture combined with appropriate optics can provide a useful balance between detector area and optical size. When thermal sensitivity below 30 mK is also available, compact cameras can support demanding inspection tasks without requiring large cooled infrared systems.
But miniaturization should not be pursued independently of measurement performance. Every reduction in sensor, lens or electronics size must still satisfy the required spatial sampling, thermal sensitivity, radiometric stability and communication bandwidth.
My Engineering View: Optimize the Measurement, Not the Specification
The most effective thermal imaging system is rarely the one with the highest number of pixels or the lowest published NETD. It is the system that delivers the required measurement with sufficient margin while remaining practical to install, trigger, communicate with and maintain.
For stationary industrial automation, I would prioritize target size, working distance, optical matching and radiometric stability before selecting the detector resolution. For embedded OEM equipment, I would additionally evaluate interface architecture, power consumption and board-level integration at an early stage.
The three-way relationship between spatial resolution, thermal sensitivity and optical cost is therefore not a limitation to be avoided. It is a design framework. Once the actual inspection requirement is understood, engineers can deliberately decide where performance is needed and where complexity can be reduced.
Industrial Applications
Compact LWIR thermography can be integrated into a range of automated processes, including:
- Continuous temperature monitoring of machinery and production equipment
- Non-contact inspection of moving workpieces
- Casting and molding process monitoring
- Thermal anomaly detection
- Electrical component and connection monitoring
- Material and process inspection
- Robotic thermal inspection
- OEM embedded vision systems
- UAV-based thermal sensing
- Condition monitoring and predictive maintenance
The final architecture may use GigE Vision and PoE for centralized factory systems or MIPI for compact embedded equipment. In either case, the design objective remains the same: obtain the required thermal information with an appropriate balance of resolution, sensitivity, optical complexity and data throughput.
