The Foundry Environment Is Not Like Other Industrial Environments
Machine vision is routinely deployed in automotive assembly, electronics manufacturing, food processing, and pharmaceutical packaging. The environmental requirements for those applications - protection against coolant spray, dust, ambient light variation - are well understood and well-served by standard industrial camera products rated IP65 or IP67.
Metal casting and forging environments add challenges that these standard ratings do not address. The combination of radiated heat from molten metal and hot parts, vibration from hydraulic presses and automated trim stations, aluminum flash and lubricant overspray, electromagnetic fields from induction billet heaters, and wide ambient temperature swings between warm foundry areas and outdoor-facing conveyor sections creates a more demanding environment than the IP rating alone characterizes.
A camera that fails in a foundry often has an intact IP67 seal. The failure mode is thermal degradation of the image sensor or lens adhesive, a vibration-induced connector failure, corrosion from alkaline lubricant residue that penetrated a gasket, or EMI-induced image corruption that looks like a software problem but traces back to insufficient shielding.
Thermal Specifications: What to Look For
Standard industrial cameras are rated for operating temperatures of 0C to 50C or sometimes 0C to 70C. This is adequate for most factory environments. On a die casting line or near a forging furnace, ambient temperatures at the camera mounting position can reach 55-65C during sustained production. This is within spec for many cameras but near the upper limit, which means that the camera is operating near its thermal derating boundary continuously - accelerating component aging.
More problematic is thermal cycling. A foundry press stops for a die change. The line goes cold. Metal production resumes and heat rises quickly. Repeated thermal cycles through a 30-40 degree range stress gaskets, expand and contract printed circuit board materials at different rates, and fatigue solder joints over time. Camera datasheets rarely specify thermal cycling endurance; you have to ask the manufacturer for this data or rely on field evidence from comparable deployments.
ForgePuls specifies cameras and compute nodes rated for -10C to 75C operating range with a thermal shock tolerance of 40C/hour change rate. The -10C lower bound handles installations near outdoor-facing areas of the plant during winter. The 75C upper bound provides margin above the typical maximum ambient near die casting machines. The thermal shock specification handles die change cycles where the line cools and reheats.
Lens selection interacts with thermal performance. Standard industrial lenses with plastic barrel components change focus position with temperature because the coefficient of thermal expansion of plastic differs from the metal lens mount and camera body. For foundry installations, metal barrel lenses with lower thermal focus shift are worth the additional cost to avoid the need for periodic refocusing during the production day.
Vibration: The Specification That Is Always Underestimated
Vibration ratings for industrial cameras are typically expressed as random vibration in g-RMS over a frequency range (e.g., 2g RMS, 10-500Hz per IEC 60068-2-64) or as sinusoidal vibration (e.g., 5g peak at specific frequencies). These ratings reflect the camera in isolation. What matters for your application is the vibration at the mounting point, which depends on mounting rigidity, the vibration source characteristics of the specific press or conveyor, and any resonance amplification in the mounting bracket.
A typical closed-die forging press generates vibration in the 5-50Hz range with peak accelerations at the press frame that can exceed 10g during the impact stroke. A camera mounted directly to the press frame with a rigid bracket will see these accelerations. The camera spec may rate it at 5g - technically out of spec for the mounting location.
The solution is vibration isolation: elastomeric mounts between the camera bracket and the press frame that attenuate high-frequency vibration before it reaches the camera. This is standard practice in metrology applications (CMMs are vibration-isolated for this reason) but is often omitted in vision system installations that copy the mounting approach from less demanding environments. Add 5-10mm rubber isolation pads or commercial anti-vibration mounts to all camera brackets on forge press installations as a baseline requirement.
Cable connections are the most vibration-sensitive component in a camera installation. Factory-terminated M12 or M8 circular connectors with locking rings are more vibration-resistant than D-Sub or RJ45 connections. Ensure cables have adequate service loops to avoid tension at the connector during press frame movement, and use tie-down points spaced at maximum 150mm intervals on cable runs in high-vibration areas.
EMI and Induction Heater Interference
Induction billet heaters, used in forging operations to heat steel or aluminum billets to forging temperature, generate strong electromagnetic fields in the kilohertz to megahertz frequency range. The field strength falls off with distance but can be significant within 2-3 meters of the heating coil. Camera image sensors and compute hardware are potential EMI victims; unshielded industrial PCs can experience processor lockups or memory errors from induction heater fields.
The ForgePuls edge compute node uses an EMI-shielded enclosure rated to IEC 61000-4 levels 3 and 4 for conducted and radiated immunity. For camera cables running near induction heaters, shielded cables with the shield terminated to ground at both ends are required. Unshielded GigE Vision cables in high-EMI zones produce image artifacts that look like random pixel noise and are often initially attributed to a camera sensor problem rather than an EMI problem.
If you are evaluating camera and compute hardware for a forging line with induction heating, request the EMI immunity test data from the hardware manufacturer - specifically the radiated immunity results per IEC 61000-4-3 at field strengths appropriate for your induction heater installation distance. If the manufacturer cannot provide this data, assume the hardware is not rated for induction heater proximity.
Lens Housing Contamination
Aluminum flash - thin fins of metal extruded from the die parting line during casting - is a hard, sharp particulate that circulates through the air in HPDC foundries. When deposited on a lens housing, flash particles scratch optical surfaces if wiped, requiring polishing or lens replacement. In high-flash environments, lens housings need either air purge (a low-volume air flow that creates positive pressure across the lens surface to prevent particle deposition) or sacrificial optical flats (replaceable protective windows in front of the lens that can be swapped without disturbing focus).
Die casting lubricant (release agent) is applied as a spray to the die cavity between shots. Overspray deposits on nearby surfaces as an oily film. On lens housings, lubricant accumulation degrades image quality within hours if not managed. Air purge systems that include a filter stage address both flash particle deposition and lubricant overspray simultaneously.
The maintenance requirement for lens housings in die casting environments - regular cleaning intervals and optical flat replacement - needs to be specified in the maintenance plan for any inspection system deployment. A maintenance interval that works in a clean assembly environment will not be adequate for a die casting line.
Specifying an Edge Compute Node for Foundry
Camera selection is only part of the hardware specification. The compute node running inference models needs to be specified for the same environment. Industrial PCs deployed in foundry environments need fanless thermal design (fans fail in dusty and contaminated air environments), wide operating temperature range, M12 or M8 circular connector Ethernet ports, and DIN-rail or panel mounting options that allow secure installation in a control cabinet near the line rather than in a distant server room.
Latency between camera capture and inspection result output is a function of both compute performance and network architecture. For in-line rejection applications where the inspection result needs to trigger an ejector or diverter mechanism, the total latency budget from camera capture to rejection signal must fit within the time the part spends in the rejection zone. At 300 shots per hour, that window is approximately 12 seconds per part. A compute node with 40-80ms inference latency running over a reliable local network provides comfortable margin.
Latency increases if inference runs on a cloud server rather than edge hardware. Network round-trip time to cloud infrastructure adds 20-80ms under normal conditions and can spike to hundreds of milliseconds during network congestion. For in-line rejection applications, edge inference is not optional. For post-process analysis and reporting, cloud connectivity is fine for aggregated data.
The hardware specification decisions outlined here are the reason ForgePuls deploys as a complete hardware and software system rather than a software-only solution. The sensor, optics, illumination, compute, and mounting hardware have been validated together for foundry environments. Assembling these components from separate vendors and integrating them in the field multiplies the points of failure and the system integration effort.
Learn about ForgePuls hardware specifications: Product Overview