Skip to content

wzyn20051216/matlab-agent-skills / skills/matlab-control-optimization

matlab-control-optimization

Verified

MATLAB workflow for control systems, system identification, robust control, MPC, Simulink Control Design, parameter tuning, constrained optimization, and numerical validation. Use whenever the user asks for controllers, transfer functions, state-space models, tuning, identification, optimal control, or solver-based engineering.

scripts/mas_ctrl_pid_design.m

View on GitHub
function r = mas_ctrl_pid_design(P, opts)
%MAS_CTRL_PID_DESIGN  PI/PID gains that MAXIMISE bandwidth subject to margins.
%
%   r = mas_ctrl_pid_design(P)                       % PI, PM>=45, GM>=6 dB
%   r = mas_ctrl_pid_design(P, struct('pmTarget',60,'gmTarget',10,'delay',Ts))
%
%   r.feasible      true if any candidate met BOTH margins
%   r.Kp, r.Ki, r.Kd
%   r.pm, r.gm, r.fcHz       margins of the chosen candidate
%   r.table         every candidate tried, with its margins
%   r.reason        why nothing was found, when feasible is false
%
% WHY A GRID AND NOT `pidtune`
% Not because a grid is better, but because the answer must be CHECKABLE. The
% candidate is graded by mas_ctrl_analyze, which reports the WORST margin over all
% 0 dB crossings; a controller that looks fine at its design crossing and has a
% second, worse one is rejected here. `pidtune` returns gains without that guarantee,
% and re-deriving the same robustness statement from its output is more work than
% sweeping, for a two-parameter loop.
%
% The objective is not "best margins" but "widest bandwidth that still meets the
% margins", because margins are a constraint, not a goal: an over-damped loop meets
% every margin and is useless.
%
% SATURATION IS NOT MODELLED HERE. A design from this function is a linear design;
% whether it survives the actuator limit has to be checked with mas_ctrl_aw_sim.

if nargin < 2, opts = struct(); end
opts = dflt(opts, struct('pmTarget', 45, 'gmTarget', 6, 'delay', 0, ...
    'zeroRatio', [0.05 0.1 0.2 0.4], 'nGain', 140, 'pid', false, ...
    'KpRange', [], 'verbose', false));

s = tf('s');
base = struct('pmTarget', opts.pmTarget, 'gmTarget', opts.gmTarget, ...
    'delay', opts.delay, 'grid', 6e4);

nr = numel(opts.zeroRatio);
if isempty(opts.KpRange)
    % start from the gain that puts the plant's own crossover near its dominant
    % pole, then sweep three decades around it. Guessing a range is unavoidable
    % without a plant model; the table makes the guess visible.
    k0 = 1/max(1e-12, abs(freqresp(P, 1)));
    opts.KpRange = logspace(log10(k0) - 2, log10(k0) + 3, opts.nGain);
end

T = struct('Kp', {}, 'Ki', {}, 'pm', {}, 'gm', {}, 'fcHz', {}, 'nCross', {}, ...
    'verdict', {}, 'ok', {});
for a = 1:numel(opts.KpRange)
    for b = 1:nr
        Kp = opts.KpRange(a);
        Ki = Kp * opts.zeroRatio(b);
        C = Kp + Ki/s;
        if opts.pid
            % derivative on the error with a fixed 1/20 of the target bandwidth as
            % the filter pole; not optimised, because a third free parameter makes
            % the grid unreadable without a real objective function
            C = C + Kp*1e-3*s/(1 + s*1e-3);
        end
        L = C*P;
        r = mas_ctrl_analyze(L, base);
        ok = (r.pmWorst >= opts.pmTarget) && (r.gmWorst >= opts.gmTarget) && ...
             strcmp(r.verdict, 'STABLE');
        T(end+1) = struct('Kp', Kp, 'Ki', Ki, 'pm', r.pmWorst, 'gm', r.gmWorst, ...
            'fcHz', pickF(r), 'nCross', r.nDown, 'verdict', r.verdict, 'ok', ok); %#ok<AGROW>
    end
end

okIdx = find([T.ok]);
r = struct();
r.table = T;
r.pmTarget = opts.pmTarget;
r.gmTarget = opts.gmTarget;
if isempty(okIdx)
    % No candidate met the constraints. Report the best margin actually achieved
    % instead of silently returning the highest-bandwidth one.
    [~, k] = max([T.pm]);
    r.feasible = false;
    r.Kp = NaN; r.Ki = NaN; r.Kd = NaN;
    r.pm = T(k).pm; r.gm = T(k).gm; r.fcHz = T(k).fcHz;
    r.reason = sprintf(['no candidate met PM>=%.0f and GM>=%.0f dB; best PM seen was ' ...
        '%.1f deg at Kp=%.4g. The plant may need a different structure (feedforward, ' ...
        'inner loop), not different gains.'], opts.pmTarget, opts.gmTarget, ...
        T(k).pm, T(k).Kp);
else
    [~, kk] = max([T(okIdx).fcHz]);
    k = okIdx(kk);
    r.feasible = true;
    r.Kp = T(k).Kp; r.Ki = T(k).Ki; r.Kd = 0;
    r.pm = T(k).pm; r.gm = T(k).gm; r.fcHz = T(k).fcHz;
    r.reason = sprintf('widest feasible bandwidth: %.4g Hz at PM %.1f deg', ...
        T(k).fcHz, T(k).pm);
end
r.margins = struct('PM', r.pm, 'GM', r.gm, 'crossings', T(k).nCross);
r.nCandidates = numel(T);
r.nFeasible = numel(okIdx);
end

% ---------------------------------------------------------------------------
function f = pickF(r)
% The reported crossover is the LOWEST downward crossing, because that is where a
% bode plot's "bandwidth" is conventionally read; the margin verdict still uses the
% worst of all crossings.
d = r.crossings(strcmp({r.crossings.kind}, 'down'));
if isempty(d), f = NaN; else, f = min([d.f]); end
end

function o = dflt(o, d)
f = fieldnames(d);
for k = 1:numel(f)
    if ~isfield(o, f{k}), o.(f{k}) = d.(f{k}); end
end
end
brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).Stars: 322kUpdated Claude CodeCodexVerifiedInstall
brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK).Stars: 198kUpdated Claude CodeCodexVerifiedInstall