Pith. sign in

REVIEW 2 major objections 5 minor 12 references

Sparse state-feedback level controllers for an industrial flotation bank can beat dense LQ on load-disturbance rejection while remaining plant-ready and easier to maintain.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-12 04:26 UTC pith:BHJXOOID

load-bearing objection Solid industrial methods paper: process-structured sparse LQ gains tuned by IAE + coordinate search beat the dense plant LQ in linear/nonlinear sims and are more maintainable; plant trial still pending. the 2 major comments →

arxiv 2607.03159 v1 pith:BHJXOOID submitted 2026-07-03 eess.SY cs.SY

Sparse State Feedback Control for Industrial Applications

Alba Gurpegui Ram\'on , Frida Norlund , Kristian Soltesz , Anders Rantzer This is my paper
classification eess.SY cs.SY
keywords sparse state-feedback controlintegral absolute errordisturbance rejectionflotation level controlindustrial MIMO controlcoordinate searchLQ comparison
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Dense linear-quadratic (LQ) feedback matrices that run on industrial plants force every actuator to depend on almost every measurement, which makes day-to-day tuning hard for operators. This paper shows how to design sparse state-feedback matrices whose non-zero pattern matches the physical neighbor interactions of a flotation bank, then tunes the free entries by coordinate search so that a weighted Integral Absolute Error (IAE) under a realistic worst-case inflow disturbance is minimized while gains stay inside plant-safe bounds and closed-loop stability is kept. Applied to the rougher flotation bank at Aitik, the resulting sparse controller improves load-disturbance rejection in the cells relative to the dense LQ already in service; the improvement appears in both linear and nonlinear simulations. Because each free parameter has a clear physical meaning and identical cells share the same gains, plant staff can adjust one interaction at a time as equipment ages. The matrices are presented as ready for deployment as a practical alternative to dense LQ.

Core claim

An optimization-based sparse state-feedback design that respects the flotation bank's interaction graph, incorporates a worst-case inflow disturbance, and minimizes a weighted IAE subject to element-wise gain bounds yields closed-loop load-disturbance rejection that is better than the dense LQ controller currently operating at Aitik, with the improvement visible in both linear and nonlinear simulations and with matrices that remain industrially interpretable and deployable.

What carries the argument

Coordinate-search optimization of the free entries of a process-structured sparse gain pair (Ks, KI,s) that minimizes the weighted IAE under a prescribed worst-case inflow disturbance while enforcing plant-derived box constraints and closed-loop stability.

Load-bearing premise

That the performance seen under the chosen linearization, the single worst-case disturbance profile, the hand-set gain bounds, and the IAE weight on the buffer will still hold on the real plant once sensor noise, valve rate limits, and unmodeled dynamics appear.

What would settle it

A side-by-side plant trial at Aitik in which the sparse matrices replace the dense LQ under a controlled milling-line stop (or equivalent inflow step) and the measured cell-level IAE fails to improve, or actuators saturate or chatter beyond the design bounds.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper proposes an optimization-based procedure for synthesizing sparse state-feedback controllers with integral action for industrial processes that admit linear control. A process-motivated sparsity mask (three central diagonals plus the first column) is imposed on the proportional and integral gains; the free nonzero entries are then tuned by a bound-constrained coordinate-search algorithm that minimizes a weighted IAE under a prescribed worst-case inflow disturbance while preserving closed-loop stability and the sign pattern of a dense LQ baseline. The method is demonstrated on the 14-state linearized model of a rougher flotation bank at the Aitik mine (already controlled by a dense LQ design). Linear and nonlinear simulations under a 50 % milling-line stop show that the optimized sparse gains reduce IAE by roughly 24 % relative to the dense LQ controller and also yield a modestly lower finite-horizon quadratic cost, while remaining within engineering gain bounds that make them candidates for plant deployment.

Significance. If the simulation gains transfer to the plant, the work supplies a practical, interpretable alternative to dense LQ for a class of industrial MIMO level-control problems. The contribution is not a new theoretical sparse-control theorem but a carefully engineered design pipeline that (i) respects process structure, (ii) optimizes a process-control metric (IAE) rather than H2, (iii) incorporates explicit gain bounds and sign constraints, and (iv) produces matrices that plant personnel can maintain. The full numerical specification of the plant model, disturbance, bounds, algorithm and resulting gains makes the study reproducible and immediately usable by practitioners facing similar flotation or cascade-tank systems.

major comments (2)
  1. The central claim of improved load-disturbance rejection (Abstract, §4, Table 1, Figs. 3–4) rests entirely on open-loop simulation of a single design disturbance (7) under fixed linearization, fixed IAE weight α=1/30 and hand-chosen element-wise bounds (8). No Monte-Carlo variation of disturbance amplitude/timing, no sensor-noise model, and no closed-loop plant data are provided. While §5.4 correctly flags the forthcoming plant trial, the manuscript currently asserts that the sparse controllers are “directly suitable for industrial deployment.” That assertion should be softened to “candidates for deployment pending plant validation,” or additional robustness checks (noise, rate limits, parameter mismatch) should be added so that the claim is supported by the evidence presented.
  2. Stability is required only as a post-hoc filter inside Algorithm 1 (“we also require that Ks(p), KI,s(p) yield a stable closed loop”). No certificate (e.g., spectral-radius bound, Lyapunov function, or even a report of the closed-loop eigenvalues of the final design) is given. Because the coordinate search can in principle leave the stable region between accepted steps, a brief verification that the returned gains keep all eigenvalues of Aaug-BaugKs in the open left half-plane (and a statement of the margin) would strengthen the safety argument for industrial use.
minor comments (5)
  1. Eq. (1) writes the nonlinear tank dynamics with f(ui) multiplying the level difference; the subsequent linearization constants (α,eta,δ,…) are given numerically but without units or a short derivation appendix. A one-sentence statement of the linearization point (h0,u0) would help reproducibility.
  2. Table 1 reports finite-horizon costs for a single initial condition x0=10·1. Adding the same metrics for the pure disturbance response of Figs. 3–4 (or for a small set of initial conditions) would make the comparison more complete.
  3. The sparsity pattern is described as “three central diagonals and the first column,” yet the optimized super-diagonal entry e is driven to zero by the algorithm. A short remark on whether this zero is retained or whether the mask could be tightened a priori would clarify the final structure.
  4. Typographical: “the imposed structure, results in gain matrices” (Abstract) needs the comma removed; “H agglund” and similar accented names appear inconsistently.
  5. Algorithm 1 is fully specified, which is excellent; a one-line statement of typical wall-clock time or number of function evaluations on the 13-parameter problem would further aid practitioners who wish to re-run the design.

Circularity Check

0 steps flagged

No significant circularity: ordinary simulation-based sparse-gain optimization evaluated on its design scenario and an independent industrial LQ baseline.

full rationale

The paper's derivation is self-contained and non-circular. The plant is linearized from first-principles mass-balance equations (1)–(2), the dense LQ baseline is obtained from the standard algebraic Riccati equation with independently chosen industrial weights (Section 3.2), and the sparsity mask is chosen from the known banded interaction topology of the flotation bank rather than from performance data. The free parameters of the sparse gains are then numerically minimized for finite-horizon IAE under a single hand-chosen disturbance (7) subject to engineering bounds (8); the same scenario is used for the numerical comparison in Table 1 and Figures 3–4. This is ordinary design-and-evaluate practice, not a self-definitional loop or a fitted quantity re-labeled as an independent prediction. The sole self-citation (Norlund et al. 2025) supplies the already-deployed LQ controller and process description; it is not invoked as a uniqueness theorem or load-bearing premise for the sparse method itself. No ansatz is smuggled via citation, no uniqueness result is imported from the authors, and no known empirical pattern is merely renamed. The central claim therefore rests on transparent numerical optimization plus simulation evidence, not on circular construction.

Axiom & Free-Parameter Ledger

5 free parameters · 4 axioms · 0 invented entities

The central performance claim rests on a linear plant model taken from prior plant work, a hand-chosen sparsity mask and gain bounds drawn from engineering experience, an ad-hoc worst-case disturbance, and an IAE weight alpha chosen to match buffer-tank behavior of the existing LQ controller. No new physical entities are postulated; the free parameters are the usual industrial tuning knobs made explicit.

free parameters (5)
  • IAE buffer weight alpha = 1/30
    Set to 1/30 so that the buffer tank exhibits the same buffering behavior under IAE optimization as under the existing LQ design; directly shapes the trade-off claimed in the results.
  • element-wise gain bounds Kb and KI,b = matrix values given in (8)
    Hand-chosen from historical plant gains and actuator limits (eq. 8); without them the optimizer produces unimplementable values. They constrain the feasible set of the claimed sparse controller.
  • worst-case disturbance amplitude and timing g(t) = -5.5e5 for 3000–7000 s
    Piecewise-constant 50 % inflow drop lasting 4000 s (eq. 7); chosen as a theoretical worst case that bypasses buffer damping. The IAE cost and all reported improvements are computed under this specific signal.
  • coordinate-search step-size schedule (gamma=0.7, initial Delta, termination 1e-4) = gamma=0.7, Delta0=max(0.1|p|,0.05 pmax)
    Heuristic parameters of Algorithm 1 that determine which local minimum is reported as p*; different schedules can yield different sparse gains.
  • LQ weighting matrices Qx, Qz, R used for baseline and initialization = diag values listed in §3.2
    Taken from the already-deployed controller (Norlund et al. 2025); they define both the dense baseline and the projection that seeds p0.
axioms (4)
  • domain assumption The nonlinear flotation dynamics are adequately captured by the given linearization (A,B) around the chosen steady state for the purpose of controller comparison.
    Stated in §2–3.1 and used for all linear simulations and for the claim that nonlinear results carry over; future plant tests are needed to confirm.
  • ad hoc to paper A three-diagonal-plus-first-column sparsity pattern is the correct process-consistent structure for the flotation bank.
    Chosen from engineering knowledge of nearest-neighbor hydraulic interactions (§3.3); different masks would produce different controllers.
  • ad hoc to paper Preserving the sign pattern of the dense LQ gains is required for interpretability and trust.
    Imposed by adjusting bounds before optimization (§3.3); not derived from stability or performance theory.
  • domain assumption Closed-loop stability is preserved by the coordinate-search procedure that only accepts improving feasible points.
    Stated as a requirement of the optimization problem (11); the algorithm itself does not contain an explicit stability certificate beyond the simulation checks.

pith-pipeline@v1.1.0-grok45 · 14602 in / 3052 out tokens · 31846 ms · 2026-07-12T04:26:23.036570+00:00 · methodology

0 comments
read the original abstract

We present an optimization-based methodology for designing sparse state-feedback controllers for industrial applications that are suited for linear control, and demonstrate the framework by designing a level controller for an industrial rougher flotation bank at the Aitik mine. In contrast to the dense linear-quadratic (LQ) controller gains currently operating at the concentrator, our approach enforces a sparsity pattern that is consistent with the interaction structure of the flotation bank and accounts for the worst-case expected inflow disturbances during tuning, while optimizing controller performance through the Integral Absolute Error (IAE) index. The non-zero elements of the sparse gain matrices are optimized using a coordinate search algorithm that handles bound constraints and preserves closed-loop stability. The resulting sparse controller achieves improved load disturbance rejection in the flotation cells compared to the LQ controller. These improvements are consistently observed in both linear and nonlinear simulations. In addition, the imposed structure, results in gain matrices that are easier to adjust and interpret. Importantly, the sparse controllers generated for the Aitik mine are directly suitable for industrial deployment and offer an effective alternative to the existing dense LQ design.

Figures

Figures reproduced from arXiv: 2607.03159 by Alba Gurpegui Ram\'on, Anders Rantzer, Frida Norlund, Kristian Soltesz.

Figure 1
Figure 1. Figure 1: A process schematic of the flotation process. [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The two parallel rougher flotation lines at the Aitik [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The inflow disturbance response of the level of the buffer tank and the levels of the flotation cells. At time 0.3 [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The inflow disturbance response of the level of the buffer tank and the froth thickness of the flotation cells. At [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

12 extracted references · 2 canonical work pages

  1. [3]

    , journal=

    Lin, Fu and Fardad, Makan and Jovanović, Mihailo R. , journal=. Design of Optimal Sparse Feedback Gains via the Alternating Direction Method of Multipliers , year=

  2. [4]

    A Splitting Method for Optimal Control , year=

    O'Donoghue, Brendan and Stathopoulos, Giorgos and Boyd, Stephen , journal=. A Splitting Method for Optimal Control , year=

  3. [5]

    , booktitle=

    Fardad, Makan and Lin, Fu and Jovanović, Mihailo R. , booktitle=. On the optimal design of structured feedback gains for interconnected systems , year=

  4. [6]

    IFAC Proceedings Volumes , volume =

    A Summary of. IFAC Proceedings Volumes , volume =. 2000 , author =

    2000

  5. [7]

    Kolda and R.M

    T.G. Kolda and R.M. Lewis and V. Torczon , title =. SIAM Review , volume =

  6. [9]

    Fardad, M., Lin, F., and Jovanović, M.R. (2009). On the optimal design of structured feedback gains for interconnected systems. In Proceedings of the 48h IEEE Conference on Decision and Control (CDC) held jointly with 2009 28th Chinese Control Conference, 978--983

    2009

  7. [10]

    and H \"a gglund, T

    Guzman, J. and H \"a gglund, T. (2024). Feedforward Control: Analysis, Design, Tuning rules, and Implementation. De Gruyter Textbook. De Gruyter, Germany. doi:10.1515/9783111429489

    work page doi:10.1515/9783111429489 2024

  8. [11]

    Kolda, T., Lewis, R., and Torczon, V. (2003). Optimization by direct search: New perspectives on some classical and modern methods. SIAM Review, 45(3), 385--482

    2003

  9. [12]

    Le Roux , J., Oosthuizen, D., Mantsho, S., and Craig, I. (2020). A survey on the status of industrial flotation control. IFAC-PapersOnLine, 53(2), 11854--11859. doi:10.1016/j.ifacol.2020.12.698

    work page doi:10.1016/j.ifacol.2020.12.698 2020

  10. [13]

    Lin, F., Fardad, M., and Jovanović, M.R. (2013). Design of optimal sparse feedback gains via the alternating direction method of multipliers. IEEE Transactions on Automatic Control, 58(9), 2426--2431

    2013

  11. [14]

    Norlund, F., Soltesz, K., and Bauer, M. (2025). Comparison of industrial controllers: Disturbance rejection in an up-and-running industrial flotation process. Journal of Process Control, 152. doi:10.1016/j.jprocont.2025.103478

    doi:10.1016/j.jprocont.2025.103478 2025

  12. [15]

    O'Donoghue, B., Stathopoulos, G., and Boyd, S. (2013). A splitting method for optimal control. IEEE Transactions on Control Systems Technology, 21(6), 2432--2442

    2013