Catch It While It's Fixable

There is a moment in module assembly after which mistakes stop being correctable. Before it, a misplaced string is a pair of tweezers and a few seconds. After it, the same misplacement is a downgrade or a scrapped module. That moment is lamination, and where inspection sits relative to it is one of the more consequential decisions in the whole line.

The sequence, and where inspection conventionally sits

The standard sequence for crystalline-silicon modules is well described. One recent study of assembly-line optimisation lays it out plainly: the line “typically includes several sequential steps: string welding, stacking, laminating, framing, and inspection”, where stacking “layers these strings with encapsulation materials such as ethylene-vinyl acetate (EVA), backsheet, and tempered glass”, the laminating step “employs high temperatures to bond the encapsulation materials with the silicon cells, forming a robust, weather-resistant composite”, and finally the modules “undergo rigorous electrical and optical performance tests, including power and electroluminescence (EL) testing”^[1].

Read that last clause again, because it is the convention this paper is arguing with. In the standard sequence, electroluminescence testing happens after lamination and framing — which is to say, after the point at which anything it finds can still be fixed.

Why lamination is the line

Lamination is not simply a hot press. It is a curing reaction: during encapsulation “two of the important material changes are the curing reaction leading to material cross-linking and interfacial adhesion formation”^[2]. Studies of the process work in the region of 125 °C to 145 °C for around ten minutes, with higher temperatures producing defects of their own — one investigation excluded 155 °C and 160 °C conditions because they “lead to large amount of bubbles”^[2]. The output is a chemically bonded stack, and that is the entire point of the step: the module has to survive decades outdoors.

The cost of that durability is that the stack cannot be taken apart again in any way that resembles rework. The evidence for this is clearest in the recycling literature, where separating a laminated module is the explicit goal and people have measured what it takes. EVA can be removed “by dissolution, thermal decomposition, and fluidized bed combustion”, and “in every case, the minimum temperature to obtain total EVA degradation is 500 °C”^[3]. Even then the cells do not all survive: the cell recovery rate, meaning cells removed without damage, “is ~85%, regardless of the method”^[3].

Undoing a lamination costs 500 °C and destroys roughly one cell in seven. That is not a rework path. It is a disposal path with a salvage rate.

We are quoting recycling research to make a manufacturing point, and it is worth being explicit that these figures come from end-of-life processing rather than factory rework. The physical constraint is the same one, and the direction of the argument does not depend on the exact number: there is no gentle way back through a cured encapsulant.

A note on the number everyone quotes

The usual way to argue this is the rule of ten — the claim that the cost of finding a defect multiplies by ten at each successive stage. We are not going to use it, and the reason is worth a paragraph, because it is a good illustration of checking a number before repeating it.

The earliest traceable published statement of the rule describes it as “possibly the most frequently quoted concept in test economics” and attributes it to two named engineers at a test equipment company “in the early seventies”, with no citation^[4]. The same author then dismantles it: “it is only a concept. It is not accurate enough to use for any analysis that may lead to a purchase decision. It was surprisingly accurate when first recognized, but this accuracy was rather short-lived”, and later, flatly, the rule “went out of the window some time ago” as component quality and mid-process test capability improved^[4].

What the peer-reviewed cost-of-quality literature supports is the directional claim, not the multiplier. The prevention–appraisal–failure model, which originates with Juran and Feigenbaum and is embedded in ASQC and BSI standards, rests on the propositions that “investment in prevention and appraisal activities will reduce failure costs, and that further investment in prevention activities will reduce appraisal costs”^[5]. That is the honest form of the argument, and our own process physics is a stronger basis for it than a borrowed heuristic anyway.

Two modalities that do not substitute for each other

Placing inspection before lamination raises an immediate question: inspect with what? The answer is both optical and electroluminescence imaging, for a reason that is easy to state and easy to get wrong. They see different physics.

Electroluminescence is band-gap emission. A current is injected and the cell emits in the near infrared, “typically in the near-infrared spectrum between 900 nm and 1300 nm… whose peak is at approximately 1150 nm”^[7]. What that reveals is electrical continuity — where current is and is not flowing. It is the only practical way to see a whole family of faults. On cell breakage, the standard field reference is blunt: “Visually it cannot be seen and in many cases it cannot be detected by a power rating of the PV module directly after occurrence of the cell breakage. Only an electroluminescence image… or a lock-in thermography image… can reveal the damage”^[6]. The same reference catalogues what EL detects: cracks that isolate cell regions from the interconnect ribbon, finger interruptions, shunt faults, disconnected interconnects, contact-forming failures, potential-induced degradation^[6].

Optical inspection sees an almost entirely different list, and the standards say so by construction. Qualification testing “require[s] an illumination of more than 1000 lux during the visual inspection and only defects detectable with the bare eye are considered”^[6]. What that catches is geometry and appearance — bubbles, delamination, discoloration, scratches, burn marks, corrosion, junction-box faults, and components “bent or misaligned” to the point of impairing installation or operation^[6].

Put those two catalogues side by side and the overlap is small. EL is largely blind to position: a perfectly conducting string in the wrong place emits beautifully. Optical inspection is largely blind to continuity: a cracked cell under an intact glass surface looks like a cell. Neither is a cheaper version of the other, and choosing between them on cost is choosing which family of defects to stop looking for.

We should be careful here about what the literature does and does not say. We have not found a source that formally establishes optical and EL inspection as complementary modalities on a production line; what exists is two well-documented and near-disjoint defect catalogues, which we have set beside each other above, plus a general call in the review literature to integrate EL “with complementary diagnostic techniques… to develop holistic PV health assessment frameworks”^[7]. The conclusion is ours, assembled from their evidence, rather than a finding we can point at.

What this means for where we put the cameras

The argument assembles into something fairly simple. Lamination is irreversible in any practical sense. Both inspection modalities see families of defects the other cannot. Several of those defect families — misplacement, contamination, a cracked cell — are trivially correctable while the stack is open and not correctable at all once it is cured. And the conventional sequence puts the electrical inspection after the point of no return.

So we intend to inspect before lamination, with both modalities, and to treat the post-lamination test as verification rather than as discovery. Finding a defect after the press is still worth doing — it is how you learn the press is drifting. But by then the module in front of you is already whatever it is going to be, and the only question left is which bin it goes in.