Teaching Scale-Up by Doing the Enterprise's Scale-Up

Process plants increasingly run reactions and separations as continuous flow through columns instead of in batches in a tank. Technicians trained only on batch labs are not prepared for what changes in that move. In a beaker, an adsorbent looks excellent. In a packed column, the same material can let pollutant slip through early, clog, or lose capacity with each regeneration. Vocational colleges rarely teach this "batch-to-continuous" gap. Chemical companies, for their part, are reluctant to let students near live continuous units, because one operator error can choke a bed or shut down a line.

This note proposes resolving both problems with one move. Student teams run small bench-scale continuous columns on campus, using a partner enterprise's real wastewater and the enterprise's own open question about whether a batch-validated process will survive continuous operation. The students learn the competencies that are specific to continuous operation. The enterprise gets scale-up data it actually needs, at no risk to its plant. The pedagogical core is a predict-then-observe loop: before running the column, each student must forecast from their own batch results how the continuous run will behave. The central measured outcome is how well students make that batch-to-continuous prediction on new, unseen chemistries.

Core Claim

We claim that the distinctive skill a continuous-process technician needs is translation: reading batch data and anticipating how it will fail in flow. This means anticipating early breakthrough, effluent plateaus, rising pressure drop, and loss of capacity across cycles. We further claim that this skill is learned best on paired batch and continuous runs of the same chemistry, rather than on batch labs alone or on idealized simulators.

The second claim concerns the enterprise side. When the student deliverable is the enterprise's own scale-up data, the enterprise's willingness to cooperate rises measurably. Willingness to cooperate is the factor that Constraints, mechanisms, and strategies for industry-education integration in vocational education identified as the most dependent bottleneck. That paper used a 534-response survey and a structural model linking twelve constraint factors. It found that enterprise willingness hinges on mutual economic interest. The proposal tests whether supplying that interest directly moves enterprise willingness, which the Shenzhen study diagnosed but did not intervene on.

What Exists and What Is Missing

The engineering literature supplies concrete cases where batch results mislead.

  • Biochar composite for chromium. A novel biochar-based 3D composite for ultrafast and selective Cr(VI) removal reports 1034 mg/g batch capacity at 323 K and 98.5% Cr(VI) removal from real leather effluent. It also reports a 32% capacity drop after the second regeneration cycle and explicitly leaves continuous-flow validation undone.
  • Copper cementation on iron spheres. Copper recovery by cementation utilizing packed bed of iron spheres shows effects that only flow produces. Flow-driven dendritic copper growth enlarged the reacting surface 2.3-fold, giving a mass-transfer correlation of $J_D = 2.6\,Re_p^{-0.5}$. Here $J_D$ is the dimensionless mass-transfer factor and $Re_p$ is the particle Reynolds number. Single-pass outlet copper plateaued near 0.1 ppm, and the paper notes short beds are prone to choking.
  • Reagent sequence in struvite precipitation. The role of reagent adding sequence in the NH4+-N recovery by MAP method shows that addition order alone moves ammonium recovery from 60–66% to 90.80% in beakers. Whether that sequence survives continuous dosing is open.

These papers motivate the proposal: they document the gap. They also supply its method: they provide the chemistries and the known outcomes against which student predictions can be scored.

The vocational and integration literature supplies institutional mechanisms, but almost entirely outside process industries.

The proposal differs from all of these in three ways:

  1. Its domain is continuous fluid processes.
  2. Its unit of instruction is a specific enterprise scale-up question.
  3. It is evaluated by a randomized comparison on a transfer test, rather than by institutional output counts.

Research Program

Partnerships. We would partner vocational colleges offering environmental or chemical process majors with nearby enterprises that hold an unresolved batch-to-continuous question. Each candidate question passes a hazard screen before it is admitted to the program. Struvite recovery and copper cementation are preferred. Cr(VI) work is admitted only where fume-hood and waste controls meet the enterprise's own standard.

Equipment. Each college builds a low-cost bench rig: glass columns of a few centimetres diameter, a peristaltic pump, an inline pressure gauge, and a fraction collector or simple online probe. This keeps equipment cost far below the AR, IoT, and blockchain stack that the field-engineer paper flagged as unaffordable for smaller institutions.

Course module. A course module of roughly one term runs in three phases:

  1. Students characterize the enterprise's effluent in batch.
  2. Each student writes a forecast for the continuous run, giving expected breakthrough time, outlet plateau, pressure-drop trend, and second-cycle capacity.
  3. Students operate the column across several shifts, compare against their forecast, and present a written scale-up recommendation to the enterprise co-mentor.

Scheduling. Continuous runs do not fit two-hour lab slots, so the module uses a rotating shift roster. This is a small-scale version of the dual-scenario alternation, and it also caps the engineer's commitment at a few review meetings. The cap directly targets the mentor time-conflict problem reported in the Greater Bay Area study.

Assessment instrument. A by-product is a reusable item bank of batch-to-continuous scenarios with known continuous outcomes. Each item gives batch data and asks for a continuous prediction. Items are seeded from the published cases above, for example:

  • predicting that the struvite sequence advantage may shrink under continuous mixing;
  • predicting the 0.1 ppm plateau in single-pass cementation;
  • predicting cycle-to-cycle capacity loss for a redox-consuming adsorbent.

The bank is extended with each enterprise run. It is the natural assessment for a stackable credential that fits the 1+X tiers: operating a column at the primary tier, diagnosing anomalies at the intermediate tier, and recommending scale-up at the advanced tier. We treat the credential as a downstream use, not the claim being tested.

Baselines

Batch-only laboratory (naive reference). This is the do-nothing reference: the existing batch laboratory curriculum on the same chemistries, with equal contact hours. It tells us how much translation skill students acquire from batch training plus lecture alone. Beating it requires a significantly higher score on held-out transfer items, which none of the arms has practised. There is no reported result found on Lacuna for transfer-test performance of batch-trained vocational students.

Virtual simulation (strongest existing method). This is the approach of the virtual teaching and research office, transposed to fluids. Students run a column simulator on the same chemistries, with breakthrough computed from standard ideal packed-bed models, and complete the same predict-then-observe loop. It is the strongest competitor because it directly attacks the equipment-cost constraint, needs no enterprise, and scales trivially. Fairness requires that it receive the same hours, the same instructor, and the same forecast-and-report structure. The only difference is that the observed outcome comes from the model rather than from a column. For outcome numbers, there is no reported result found on Lacuna for learning outcomes of simulation-based process training. The nearest reported figure is the 42% shorter job adaptation cycle in the field-engineer system. That figure comes from a different domain, comes from a design coupled to physical equipment, and has no described control group, so we treat it as context rather than as a target. Beating this arm means outperforming it on items whose true outcome departs from ideal behaviour, such as clogging, plateaus, and cycle loss. On items where ideal behaviour holds, it should be matched, not necessarily beaten.

Generic physical column (ablation of the key ingredient). This arm has students run the same bench rig on a standard teaching chemistry, copper cementation on iron spheres, with no enterprise involvement. It isolates what the enterprise problem adds beyond physical continuous operation. All three non-naive arms share the forecast step. A within-arm variant of the full proposal drops the forecast step, which isolates the predict-then-observe loop itself.

Why not just…?

Why not just teach batch fundamentals and let continuous operation be learned on the job? The cited engineering papers show that batch results can be confidently wrong about flow behaviour: the 32% cycle loss, the 0.1 ppm plateau, and flow-induced surface change. On the job, those lessons are learned on live equipment, which is exactly the risk that keeps chemical enterprises closed to students. The batch-only arm measures how large the translation deficit actually is. If that arm scores close to the others on held-out items, the proposal's premise fails and we would report that.

Why not just use simulation software and skip physical columns? This is the most serious threat on student outcomes. Simulators capture ideal breakthrough well. Whether students also need to experience non-ideal failures is an empirical question, and the item bank is deliberately split into ideal-behaviour items and non-ideal items to answer it. If the simulation arm matches the physical arms on non-ideal items too, the recommendation becomes simulation plus enterprise data review, and the rig is dropped.

Why not just a cheap generic column with no enterprise partner? This is also a real threat. If it matches the full proposal on student outcomes, then the enterprise ingredient earns its place only through enterprise engagement, and we would say so. That is why enterprise-side outcomes are measured separately rather than folded into a single score.

Why not just conventional end-of-program internships? The Shenzhen study and the dual-scenario study both describe such placements as observational and weakly aligned with enterprise interest. More decisively, process enterprises rarely grant hands-on access to live continuous units, so an internship cannot deliver the practice we are trying to provide. The enterprise-engagement measures below test whether our alternative actually changes enterprise behaviour.

Proposed Experiments

Design. The main study is a cluster-randomized comparison. Classes within participating colleges are assigned to one of four conditions: batch-only, simulation, generic column, or enterprise-problem column. Within the enterprise-problem arm, a further split separates students with and without the forecast step. Randomizing classes rather than individuals avoids contamination within a lab group. The analysis is a multilevel model of transfer score with class and college as random effects. Enough classes must be recruited per arm to detect a moderate effect at the cluster level, and we would report the achieved minimum detectable effect rather than overclaim.

Student outcomes.

  • Primary outcome: accuracy on held-out transfer items involving chemistries none of the arms trained on, reported separately for ideal-behaviour and non-ideal items.
  • Hands-on practical: bringing a column to steady state and responding to an injected disturbance, such as a pressure rise, scored on a rubric by enterprise assessors blinded to arm.
  • Follow-up: a six-month follow-up of graduates who enter process employers, recording supervisor-rated time to independent operation. This addresses the longitudinal gap the integration literature repeatedly flags.

Enterprise outcomes.

  • Re-administer the Shenzhen constraint items on willingness to cooperate, mutual interest, and teacher involvement to partner enterprise staff before and after the module.
  • Record whether each enterprise renews for a second term.
  • Record whether it used the student scale-up data in an internal decision.
  • Log engineer hours, to test whether bounded review roles sustain participation better than the co-teaching load reported in the computing professional group.

Decisive checks. The proposal succeeds if two results hold together:

  1. The physical arms beat the simulation arm specifically on non-ideal transfer items.
  2. The enterprise-problem arm raises enterprise renewal and the cooperation-willingness items relative to the generic column, without lowering student scores.

If only the first holds, the contribution is a low-cost physical module, not an integration mechanism. If only the second holds, the contribution is a partnership device, not a pedagogy. Either partial result is still informative.