Industry engagement with control research: Perspective and messages

Authors: Tariq Samad, Margret Bauer, Scott Bortoff, Stefano Di Cairano, Lorenzo Fagiano, Peter Fogh Odgaard, R. Russell Rhinehart, Ricardo Sánchez-Peña, Atanas Serbezov, Finn Ankersen, Philippe Goupil, Benyamin Grosman, Marcel Heertjes, Iven Mareels, Raye Sosseh Year: 2020 Tags: control-engineering, theory-practice-gap, industry-academia-engagement, technology-transfer, process-control, control-education

TL;DR

The IFAC Industry Committee argues that the control theory-practice gap is not about field vitality—patents and publications are growing robustly—but about collapsing mutual engagement between researchers and practitioners. Ten structured "messages" for researchers are backed by survey data and participation metrics, aiming to convey the industrial mindset on priorities, constraints, and incentives.

First pass — the five C's

Category. Position/perspective article with supporting survey data; not a research prototype or empirical study.

Context. Control systems engineering, theory-practice gap discourse. Builds on Foss (1973) — earliest cited articulation of the gap in process control; Bernstein et al. (1999) — IEEE CSM special section on theory-practice gap; Qin & Badgwell (2003) — MPC industrial deployment census; Åström & Kumar (2014) — broad historical review of control applications.

Correctness. Three load-bearing assumptions: (1) IFAC/IEEE-CSS participation is a valid proxy for industry-academia engagement overall; (2) IFAC Industry Committee survey respondents represent industry broadly; (3) patent counts containing "control system" reliably track industry investment in control innovation. None of the three is validated in the paper.

Contributions. - Quantifies the participation collapse: IEEE-CSS industry leadership share fell from a majority in 1961–1970 to 0% from 2011 onward; IFAC industry attendance dropped from 19% (2000–2002) to 14% (2012–2014). - Presents a 66-respondent impact survey showing MPC rated high-impact by 62% (current) and 85% (future), versus robust control 26%/42%, adaptive control 18%/44%, game theory 5%/17%. - Documents cost reduction as the top-ranked innovation driver in 7 of 8 industry sectors surveyed, directly contradicting academia's near-exclusive focus on performance metrics. - Formulates ten practitioner-derived "messages" covering domain expertise, implementation infrastructure, economic constraints, and education gaps.

Clarity. Readable and well-structured overall; the ten messages vary substantially in evidentiary support — some are grounded in survey data, others rest entirely on committee anecdote — and the paper ends mid-sentence due to a truncated excerpt.

Second pass — content

Main thrust: Industry is investing in control innovation (control-related patents grew 7.60× from 1998–2017 vs. 3.17× for all patents globally), but industry voices have nearly vanished from research organizations, and researchers systematically misread industry priorities — especially underweighting cost reduction and overweighting theoretical elegance.

Supporting evidence: - Control-related patents grew 7.60× (1998–2017) vs. 3.17× total; in China, 130× vs. 38× total — control patents grew ~3× faster than average. - IEEE-CSS: zero industry-affiliated leaders from 2011 onward; absolute industry leader count also declined despite total leadership positions growing. - 2018 survey (n=66 of 77 committee members): PID rated high-impact by 91%; MPC 62%; system ID 65%; nonlinear control 21%; hybrid systems 11%. - Sector-specific key-driver survey: cost reduction ranks first in aerospace (n=5), energy/oil & gas (n=13), and process industry (n=31); performance rarely tops the list. - Market data (Table 5): industrial control CAGR 5.3% vs. digital transformation 18.2% and IoT 24.7% (2018 base year, commercial market report sources).

Figures & tables: Fig. 1 (IFAC congress paper counts) and Fig. 2 (patent trends) have labeled axes but no error bars or significance tests — both are descriptive counts. Fig. 3(a,b) show IEEE-CSS leadership trends by decade; no uncertainty quantification. Table 2 reports survey percentages with no confidence intervals. Table 3 sector breakdowns have dangerously small sub-group n (aerospace=14, automotive=10). Table 4 sector drivers include groups as small as n=2 (robotics) and n=3 (medical). Table 5 market figures are sourced exclusively from commercial market reports via tinyurl links with no independent verification — a significant reliability concern.

Follow-up references: - Åström & Kumar (2014) — comprehensive historical account of control applications worth reading alongside this piece. - Qin & Badgwell (2003) — primary source for MPC industrial deployment figures cited repeatedly here. - Samad & Annaswamy (2014) — compendium of control success stories referenced as the broadest available catalog. - Rossiter et al. (2019) — pilot survey on curriculum priorities as seen by industry vs. academia, the direct precursor to the education workstream described in §4.10.

Third pass — critique

Implicit assumptions: - IFAC/IEEE-CSS participation is the right measure of industry-academia engagement; if industry engages primarily through bilateral R&D contracts, consortia, or secondments, the metric systematically misses it and the "problem" may be overstated. - Committee-member survey responses generalize to industry at large; this is strongly confounded by self-selection — IFAC members are atypical practitioners. - Patent keyword search on "control system" is a valid measure; the string is broad enough to capture patents where control is incidental and narrow enough to miss sector-specific vocabulary (e.g., "autopilot," "governor," "regulator"). - The theory-practice gap is a structural problem requiring community-level intervention rather than a natural and functional division of labor.

Missing context or citations: - No evidence is provided that greater academic-industry engagement actually improves deployed system performance — the normative claim that engagement is beneficial is asserted, not demonstrated. - No comparison with analogous engineering subfields (signal processing, power electronics, ML) to validate that the gap is "especially prominent" in control specifically. - Table 5 market figures rely entirely on commercial market reports with no academic or government corroboration and no methodology for CAGR estimates. - The paper acknowledges its own 2019 large-scale education survey results are not yet available, making §4.10 substantially incomplete.

Possible experimental / analytical issues: - No statistical tests anywhere — all percentage comparisons in Tables 2–4 lack confidence intervals or p-values; sub-group sizes in Table 3 (n=10–34) and Table 4 (n=2–31) make sector-level inferences unreliable. - IEEE-CSS leadership analysis conflates proportional and absolute trends without resolving an internal inconsistency: the paper states total positions grew yet industry absolute numbers also fell — the denominator behavior is not clearly shown in the figures. - The ten messages are not systematically derived from the survey data; at least half rest on informal committee consensus, which is not distinguished from empirical findings. - Success story vignettes in §5 are qualitative and selected for impact — no failed-transfer cases are discussed, creating survivorship bias in the overall argument. - The excerpt is truncated (ends mid-sentence in §5.2), so the complete success stories section and conclusion cannot be evaluated.

Ideas for future work: - Cross-field comparison: replicate the IEEE-CSS leadership and conference attendance analysis for IEEE Signal Processing Society and IEEE Power Electronics Society to test whether the engagement decline is control-specific or field-agnostic. - Unbiased industry survey: repeat the impact and priority surveys with a random sample of practicing engineers identified through professional licensing databases rather than IFAC membership rolls. - Longitudinal outcome study: track whether IFAC Industry Committee interventions post-2017 produce measurable changes in industry participation metrics over a 5–10 year window, providing a natural quasi-experiment. - Transfer-failure analysis: systematically document cases where advanced control algorithms were evaluated industrially but not deployed, to characterize which of the ten identified barriers were decisive.