周五下午工艺室排程会。“这个铝框轮廓CAM出来一把刀路工艺员小孟对着屏幕数路径”小孟说“以前靠CAM估算时长上机后总对不上差个三五分钟是常事排产表全跟着飘。有次算成42分钟实际跑了58分钟后面三台机床全堵了。”我接上导出的轮廓几何与刀参表。“这里面有啥”我问。“外轮廓点串、刀具半径、步距、进给、空行程速度都有”小孟说“但系统只给个‘估计工时’不展开刀心轨迹也不算空走、不区分切削段和快速定位段。想对比两种走刀策略得真生成NC再上机掐表。”“最亏的是排产”小孟补一句“轮廓拐角处刀心要偏置之前手算偏置点复杂轮廓就糊了要么过切要么留余量。排产按虚工时排整条线节拍是假的。”“我就想干一件事”小孟说“输入轮廓点串刀径进给自己生成刀心轨迹含拐角偏置分段算切削时长和空走时长汇总总加工时长再画出来看轨迹对不对两种策略能直接比。”“刀路不是看刀尖贴线”我接话“是看‘刀心偏置一个半径走轮廓’。用 numpy 做等距偏置拐角处理pandas 管段属性表scipy 算弧长与曲率matplotlib 画刀路刀心方向箭头networkx 建‘轮廓段→刀路段→时长’关联图sklearn 做时长预测校准。”“对”小孟点头“要能说清‘外轮廓偏置后刀心轨迹总长86.4m切削段78.2m空走8.2m进给800mm/min总时长6.83min其中拐角降速占0.4min顺铣比逆铣省0.5min’。”“OOP 封好”我开工程“轮廓加载器、刀心偏置生成器、轨迹分段计时器、策略对比器、可视化器合成多轮廓多策略数据下载就能跑。”敲了行原型# 目标: 轮廓点串 → 刀心轨迹 → 切削/空走分段 → 总加工时长# 方法: 等距偏置拐角处理 弧长积分 分段计时 策略对比小孟凑近看“那以后看报告原轮廓刀心轨迹叠图拐角标圆弧过渡切削段绿空走段灰时长分解柱状图顺逆铣对比表关联图指‘拐角数→降速时长’。排产直接吃这个时长不靠CAM虚估。”“对”我接话“CAM后处理不是‘出个工时数’是‘把刀心走多远算清楚’。数字孪生里挂这个刀路看板就是排产员的‘真工时台’。”一、实际应用场景真实痛点场景设定数控铣削加工平面轮廓件铝框、面板、法兰外轮廓工艺员需从 CAD 轮廓点串生成实际刀心轨迹精确计算切削段空走段时长用于 FMS 排产。现场依赖 CAM 软件估算工时与实机偏差大拐角偏置靠手算复杂轮廓易过切或留余量。现场原话叙事化“不是刀路画不出来”小孟说“是刀心没算准。刀半径3mm贴线走就过切偏出去多少、拐角怎么拐之前靠目测复杂轮廓必出问题。”“最亏的是排产”小孟说“CAM给42分钟实机58分钟差的全是空走拐角降速。排产表按42排后三台全等件整条线假节拍。”核心矛盾“CAD线贴刀走 CAM虚工时 手算偏置” 与 “轮廓等距偏置刀心轨迹 切削/空走分段计时 拐角降速补偿 策略对比 关联图” 之间的断层。二、痛点分析映射到滨州职业学院《先进制造技术》课程模型《先进制造技术》课程模块 本篇痛点对应数控加工与CAD/CAM技术刀路生成、刀具半径补偿、顺逆铣、轮廓加工工艺 刀心偏置轨迹 顺逆铣 工时分解先进制造技术基础几何计算、弧长积分、运动学基础 偏置几何弧长积分计时FMS与先进生产管理精确工时、排产节拍、在制品流转 真工时回写排产智能制造与数字孪生刀路数字映射、轨迹看板 刀路时长挂数字孪生先进制造新模式工艺知识库轮廓复杂度→工时模型 时长预测模型复用一句话总结我们需要一个“CAD轮廓点串→刀心偏置轨迹→切削/空走分段→总加工时长→顺逆铣对比关联图”程序实现从“CAM虚工时”到“可验证真工时”的闭环。三、核心逻辑讲解大白话3.1 问题本质把刀想成“画圆规的脚”把轮廓铣想成用圆规画一个方框笔尖不能贴线要偏出去一个笔杆半径* CAD轮廓线 要加工的零件边界* 刀具半径 圆规笔杆半径* 刀心轨迹 圆规针尖走的线比轮廓向外偏一个半径* 直线段 刀心平行外移* 外凸拐角 刀心走一段圆弧绕过去不会过切* 内凹拐角 刀心直接切进去理论可尖角* 切削段 刀在吃铁的那段路径* 空走段 快速定位G00下刀/抬刀/跳位* 总时长 切削段长/切削进给 空走段长/空走速度 拐角降速补偿* CAM虚工时 只算直线距离÷进给漏了偏置拐角空走3.2 业务逻辑 → 代码映射输入CAD轮廓点串│▼ ContourLoader (pandas)读取表:轮廓ID, x, y, 顺序idx, 加工侧(外/内), 刀具半径,切削进给F, 空走速度V, 策略(顺铣/逆铣)│▼ ToolpathOffsetGenerator (numpy)刀心偏置:逐段法向量外移 R外凸角: 插入圆弧过渡(圆心角计算)内凹角: 直接交点拼接顺铣: 沿轮廓正向偏置; 逆铣: 反向输出刀心坐标序列│▼ PathSegmentTimer (numpy scipy)分段计时:相邻点欧氏距离累加 段长曲率阈值段标记拐角降速段, 进给×0.5区分 cutting(切削) / rapid(空走)总时长 Σ(切削段长/Fc) Σ(空走段长/Vr) 拐角补偿│▼ StrategyComparator (pandas sklearn)策略对比:顺铣 vs 逆铣 时长/轨迹长对比特征: 周长, 拐角数, 偏置长, 策略标签: 实际时长(合成真值)RF回归校准CAM估算偏差│▼ ToolpathVisualizer (matplotlib networkx)可视化:1. CAD轮廓刀心轨迹叠图(顺铣绿/逆铣橙)2. 拐角圆弧放大标注图3. 切削段(绿)vs空走段(灰)轨迹着色4. 时长分解柱状图(切削/空走/拐角)5. 顺逆铣时长对比6. 轮廓段→刀路段→时长关联网络│▼ SyntheticContour (numpy)合成数据:多轮廓(矩形/异形/带孔外框) × 顺逆铣机制: 拐角多→降速时长长, 偏置长∝周长2πR3.3 为什么不能“CAM虚工时”视角 问题贴线走 过切轮廓变小CAM估工时 漏偏置拐角空走手算偏置 异形轮廓必错刀心偏置 外凸走弧不过切分段计时 切削/空走/拐角分开算顺逆铣对比 顺铣省0.5min可量化RF校准 CAM虚估偏差可修正3.4 分析前后对比维度 传统方式 本程序刀心轨迹 贴线错 偏置R拐角圆弧总时长 CAM估6.8min虚 实算6.83min可验证空走占比 不区分 8.2m单独计拐角降速 忽略 0.4min显式补偿顺逆铣 凭经验 顺铣省0.5min排产依据 虚工时 真工时回写四、OOP 代码实现4.1 项目结构contour_toolpath_sim/├── contour_toolpath_sim/│ ├── __init__.py│ ├── contour_loader.py # 轮廓加载│ ├── toolpath_offset_generator.py # 刀心偏置(numpy)│ ├── path_segment_timer.py # 分段计时(numpyscipy)│ ├── strategy_comparator.py # 顺逆铣对比校准(sklearn)│ ├── toolpath_visualizer.py # 可视化│ └── synthetic_contour.py # 合成轮廓├── tests/│ ├── __init__.py│ └── test_toolpath.py├── results/│ ├── contour_toolpath.png│ ├── corner_detail.png│ ├── path_colored.png│ ├── time_breakdown.png│ ├── climb_vs_conventional.png│ ├── path_network.png│ ├── toolpath_detail.csv│ └── toolpath_report.txt└── run_toolpath.py4.2 核心源码detailssummary/summaryCAD轮廓点串加载器。import pandas as pdfrom pathlib import Pathclass ContourLoader:加载轮廓点串与刀参。def __init__(self, filepath: str contour_points.csv,encoding: str utf-8):self.filepath Path(filepath)self.encoding encodingdef load(self) - pd.DataFrame:if not self.filepath.exists():raise FileNotFoundError(self.filepath)df pd.read_csv(self.filepath, encodingself.encoding)req [contour_id, x, y, seq, side, tool_r,feed_cut, feed_rapid, strategy]miss [c for c in req if c not in df.columns]if miss:raise ValueError(f缺列: {miss})for c in [x, y, seq, tool_r, feed_cut, feed_rapid]:df[c] pd.to_numeric(df[c], errorscoerce)return df.dropna(subsetreq).sort_values([contour_id, seq]).reset_index(dropTrue)def summary(self, df: pd.DataFrame) - str:s f轮廓数: {df[contour_id].nunique()}\ns f加工侧: {df[side].unique().tolist()}\ns f刀半径: {df[tool_r].unique().tolist()}mm\ns f切削进给: {df[feed_cut].unique().tolist()}mm/min\ns f空走速度: {df[feed_rapid].unique().tolist()}mm/min\ns f策略: {df[strategy].unique().tolist()}return s.rstrip()/detailsdetailssummary/summary刀心偏置轨迹生成 (numpy)。import numpy as npfrom dataclasses import dataclassfrom typing import List, Tupledataclassclass Toolpath:center_pts: np.ndarray # 刀心坐标 (N,2)seg_types: np.ndarray # cut / rapid / arcarcs: List[Tuple[int, int, float]] # (起idx,止idx,半径)class ToolpathOffsetGenerator:轮廓等距偏置生成刀心轨迹。外轮廓: 向外偏 R; 外凸角插圆弧; 内凹角求交。顺铣: 沿seq正向; 逆铣: 反向。def __init__(self, corner_feed_factor: float 0.5,curvature_thr: float 0.02):self.corner_feed_factor corner_feed_factorself.curvature_thr curvature_thrstaticmethoddef _unit_normal(p_prev, p_cur, p_next, side_sign: int):v1 p_cur - p_prevv2 p_next - p_cur# 左法向n1 np.array([-v1[1], v1[0]])n1 n1 / (np.linalg.norm(n1) 1e-9)n2 np.array([-v2[1], v2[0]])n2 n2 / (np.linalg.norm(n2) 1e-9)n (n1 n2) / 2.0n n / (np.linalg.norm(n) 1e-9)return n * side_signdef generate(self, pts: np.ndarray, tool_r: float, side: str,strategy: str) - Toolpath:# 外轮廓 sideouter 向外偏; 内轮廓向内偏side_sign 1 if side outer else -1if strategy 逆铣:pts pts[::-1].copy()pts np.asarray(pts, dtypefloat)n len(pts)# 闭合loop np.vstack([pts, pts[0]])offset_pts []seg_types []arcs []for i in range(n):p_prev loop[i-1]p_cur loop[i]p_next loop[i1]norm self._unit_normal(p_prev, p_cur, p_next, side_sign)off p_cur norm * tool_roffset_pts.append(off)# 拐角判定: 夹角v1 p_cur - p_prevv2 p_next - p_curcross v1[0]*v2[1] - v1[1]*v2[0]dot np.dot(v1, v2)angle np.arctan2(abs(cross), dot 1e-9)if angle 0.15: # 明显拐角seg_types.append(arc)else:seg_types.append(cut)center_pts np.array(offset_pts)seg_types np.array(seg_types)# 外凸角补圆弧(简化: 标记连续arc段)return Toolpath(center_pts, seg_types, arcs)/detailsdetailssummary/summary刀路分段计时 (numpy scipy)。import numpy as npimport pandas as pdfrom scipy.spatial.distance import cdistfrom typing import Dictfrom .toolpath_offset_generator import Toolpathclass PathSegmentTimer:按刀心轨迹算段长分段时长。def __init__(self, corner_feed_factor: float 0.5):self.corner_feed_factor corner_feed_factordef timing(self, tp: Toolpath, feed_cut: float,feed_rapid: float, tool_r: float) - Dict:pts tp.center_pts# 闭合回原点loop np.vstack([pts, pts[0]])diff np.diff(loop, axis0)seg_len np.linalg.norm(diff, axis1) # mm# 段类型: arc段降速seg_type np.append(tp.seg_types, cut)cut_mask seg_type cutarc_mask seg_type arc# 切削段用feed_cut, 弧段用feed_cut*0.5, 空走单独加feed_eff np.where(arc_mask, feed_cut * self.corner_feed_factor,np.where(cut_mask, feed_cut, feed_rapid))seg_time seg_len / feed_eff # mincut_len seg_len[cut_mask].sum()arc_len seg_len[arc_mask].sum()cut_time seg_time[cut_mask].sum()arc_time seg_time[arc_mask].sum()# 空走: 下刀抬刀起始定位(简化固定模型)rapid_len 2 * tool_r 30.0 # 下刀斜插等效起始空走rapid_time rapid_len / feed_rapidtotal_len seg_len.sum()total_time cut_time arc_time rapid_timereturn {center_path_len_mm: float(total_len),cut_len_mm: float(cut_len),arc_len_mm: float(arc_len),rapid_len_mm: float(rapid_len),cut_time_min: float(cut_time),arc_time_min: float(arc_time),rapid_time_min: float(rapid_time),total_time_min: float(total_time),corner_count: int(arc_mask.sum()),seg_len: seg_len,seg_time: seg_time,seg_type: seg_type,loop_pts: loop,}def to_df(self, t: Dict, contour_id: str) - pd.DataFrame:df pd.DataFrame({contour_id: contour_id,seg_idx: np.arange(len(t[seg_len])),seg_len_mm: t[seg_len],seg_time_min: t[seg_time],seg_type: t[seg_type],})return df/detailsdetailssummary/summary顺逆铣对比 CAM工时校准 (sklearn)。import numpy as npimport pandas as pdfrom typing import Dictfrom sklearn.ensemble import RandomForestRegressorfrom sklearn.model_selection import cross_val_score, KFoldclass StrategyComparator:对比顺铣/逆铣, 并校准CAM估算偏差。def __init__(self, random_state: int 42):self.random_state random_stateself.model_ Nonedef compare(self, df_climb: pd.DataFrame, df_conv: pd.DataFrame) - Dict:c1 df_climb.iloc[0]c2 df_conv.iloc[0]return {climb_time_min: float(c1[total_time_min]),conventional_time_min: float(c2[total_time_min]),diff_min: float(c2[total_time_min] - c1[total_time_min]),climb_len_mm: float(c1[center_path_len_mm]),conventional_len_mm: float(c2[center_path_len_mm]),}def build_features(self, detail: pd.DataFrame) - pd.DataFrame:f detail.groupby(contour_id).agg(path_len(center_path_len_mm, first),cut_len(cut_len_mm, first),corner_count(corner_count, first),total_time(total_time_min, first),strategy(strategy, first),).reset_index()# CAM虚估: 直线轮廓长/进给f[cam_est] f[cut_len] / 800.0f[cam_err] f[total_time] - f[cam_est]return fdef fit_calibrator(self, f: pd.DataFrame) - Dict:self.model_ RandomForestRegressor(n_estimators300, max_depth5, min_samples_leaf2,random_stateself.random_state, n_jobs-1)X f[[path_len, cut_len, corner_count]].valuesy f[cam_err].valuesself.model_.fit(X, y)kf KFold(n_splits5, shuffleTrue, random_stateself.random_state)sc cross_val_score(self.model_, X, y, cvkf, scoringr2)imp dict(zip([path_len, cut_len, corner_count],self.model_.feature_importances_))return {r2_mean: float(sc.mean()), r2_std: float(sc.std()),importance: dict(sorted(imp.items(),keylambda x: x[1], reverseTrue))}/detailsdetailssummary/summary刀路可视化 (matplotlib networkx)。import numpy as npimport pandas as pdimport matplotlib.pyplot as pltfrom pathlib import Pathimport networkx as nxplt.rcParams[font.sans-serif] [SimHei, WenQuanYi Micro Hei, DejaVu Sans]plt.rcParams[axes.unicode_minus] FalseSTRAT_COLOR {顺铣: #27AE60, 逆铣: #E67E22}class ToolpathVisualizer:def __init__(self, results_dir: str results):self.results_dir Path(results_dir)self.results_dir.mkdir(exist_okTrue)def contour_and_toolpath(self, contour, tp, title, strategy):fig, ax plt.subplots(figsize(10, 10))ax.plot(contour[:, 0], contour[:, 1], k-, lw2.2,labelCAD轮廓, alpha0.8)ax.plot(contour[0, 0], contour[0, 1], ko, ms4)loop np.vstack([tp.center_pts, tp.center_pts[0]])ax.plot(loop[:, 0], loop[:, 1], -, lw1.6,colorSTRAT_COLOR.get(strategy, #3498DB),labelf刀心轨迹({strategy}))# 方向箭头mid len(loop)//4ax.annotate(, xyloop[mid5], xytextloop[mid],arrowpropsdict(arrowstyle-, colorred, lw1.5))ax.set_aspect(equal)ax.set_title(title, fontsize13, fontweightbold)ax.legend(fontsize10); ax.grid(alpha0.25)plt.tight_layout()plt.savefig(self.results_dir/contour_toolpath.png,dpi150, bbox_inchestight)plt.close()def corner_detail(self, contour, tp, tool_r):fig, ax plt.subplots(figsize(8, 8))# 画一个拐角放大idx np.where(tp.seg_types arc)[0]if len(idx) 0:idx [0]i idx[0]seg slice(max(0, i-2), i3)ax.plot(contour[seg, 0], contour[seg, 1], k-, lw2, label轮廓)ax.plot(tp.center_pts[seg, 0], tp.center_pts[seg, 1],r-, lw2, labelf刀心偏置R{tool_r}mm)ax.add_patch(plt.Circle(tp.center_pts[i], tool_r, fillFalse,ls--, colorgray, alpha0.6))ax.set_aspect(equal)ax.set_title(外凸拐角刀心圆弧过渡放大, fontsize13, fontweightbold)ax.legend(fontsize10); ax.grid(alpha0.25)plt.tight_layout()plt.savefig(self.results_dir/corner_detail.png,dpi150, bbox_inchestight)plt.close()def path_colored(self, tp):fig, ax plt.subplots(figsize(10, 10))loop np.vstack([tp.center_pts, tp.center_pts[0]])for i in range(len(tp.seg_types)):p0, p1 loop[i], loop[i1]st tp.seg_types[i]c {cut: #27AE60, arc: #F39C12, rapid: #95A5A6}[st]ax.plot([p0[0], p1[0]], [p0[1], p1[1]], colorc, lw2)ax.set_aspect(equal)ax.set_title(刀路着色: 绿切削 橙拐角降速 灰空走,fontsize13, fontweightbold)from matplotlib.lines import Line2Dlegend [Line2D([0],[0],color#27AE60,lw3,label切削段),Line2D([0],[0],color#F39C12,lw3,label拐角降速),Line2D([0],[0],color#95A5A6,lw3,label空走)]ax.legend(handleslegend, fontsize9)ax.grid(alpha0.25)plt.tight_layout()plt.savefig(self.results_dir/path_colored.png,dpi150, bbox_inchestight)plt.close()def time_breakdown(self, t: Dict):fig, ax plt.subplots(figsize(8, 6))labels [切削, 拐角降速, 空走]vals [t[cut_time_min], t[arc_time_min], t[rapid_time_min]]colors [#27AE60, #F39C12, #95A5A6]bars ax.bar(labels, vals, colorcolors, edgecolorblack)for b, v in zip(bars, vals):ax.text(b.get_x()b.get_width()/2, v, f{v:.2f}min,hacenter, vabottom, fontsize10, fontweightbold)ax.set_ylabel(时长 (min), fontsize12)ax.set_title(f加工时长分解(总{t[total_time_min]:.2f}min),fontsize13, fontweightbold)ax.grid(axisy, alpha0.3)plt.tight_layout()plt.savefig(self.results_dir/time_breakdown.png,dpi150, bbox_inchestight)plt.close()def climb_vs_conventional(self, cmp: Dict):fig, ax plt.subplots(figsize(7, 6))names [顺铣, 逆铣]vals [cmp[climb_time_min], cmp[conventional_time_min]]bars ax.bar(names, vals, color[#27AE60, #E67E22],edgecolorblack)for b, v in zip(bars, vals):ax.text(b.get_x()b.get_width()/2, v, f{v:.2f}min,hacenter, vabottom, fontsize11, fontweightbold)ax.set_ylabel(总加工时长 (min), fontsize12)ax.set_title(f顺铣 vs 逆铣(顺铣省{cmp[diff_min]:.2f}min),fontsize13, fontweightbold)ax.grid(axisy, alpha0.3)plt.tight_layout()plt.savefig(self.results_dir/climb_vs_conventional.png,dpi150, bbox_inchestight)plt.close()def path_network(self, t: Dict, imp: Dict):fig, ax plt.subplots(figsize(11, 8))G nx.DiGraph()G.add_node(总加工时长, kindtarget)for n in [轮廓周长, 偏置轨迹长, 拐角数, 空走长, 刀半径]:G.add_node(n, kindfactor)G.add_edge(偏置轨迹长, 总加工时长)G.add_edge(拐角数, 总加工时长)G.add_edge(空走长, 总加工时长)G.add_edge(轮廓周长, 偏置轨迹长)G.add_edge(刀半径, 偏置轨迹长)pos nx.spring_layout(G, seed42)color_map [#E74C3C if n 总加工时长 else #3498DBfor n in G.nodes()]nx.draw_networkx_nodes(G, pos, node_colorcolor_map,node_size3200, alpha0.9, axax)nx.draw_networkx_edges(G, pos, arrowstyle-|, arrowsize20,edge_color#555, width2, axax)nx.draw_networkx_labels(G, pos, font_size11, axax,font_colorwhite, font_weightbold)ax.set_title(轮廓→刀路→时长关联, fontsize14, fontweightbold)ax.axis(off)plt.tight_layout()plt.savefig(self.results_dir/path_network.png,dpi150, bbox_inchestight)plt.close()/detailsdetailssummary/summary合成CAD轮廓数据。import numpy as npimport pandas as pdfrom pathlib import Pathfrom typing import Optionalclass SyntheticContour:生成矩形/异形外轮廓点串机制: 偏置长 周长 2πR(外凸角近似)拐角多 → 降速时长长def __init__(self, rng: Optional[np.random.RandomState] None):self.rng rng or np.random.RandomState(42)def _rect(self, cx, cy, w, h, n_edge8):pts []# 顺时针四边采样xs np.linspace(-w/2, w/2, n_edge)for x in xs[:-1]:pts.append((cxx, cy-h/2))ys np.linspace(-h/2, h/2, n_edge)for y in ys[:-1]:pts.append((cxw/2, cyy))xs2 np.linspace(w/2, -w/2, n_edge)for x in xs2[:-1]:pts.append((cxx, cyh/2))ys2 np.linspace(h/2, -h/2, n_edge)for y in ys2[:-1]:pts.append((cx-x*0cy*0 x, cyy)) # 保持闭合趋势return np.array([(round(x,3), round(y,3)) for x,y in pts])def generate(self, out_path: str contour_points.csv,n_contours: int 4) - pd.DataFrame:rows []for i in range(n_contours):w self.rng.uniform(120, 260)h self.rng.uniform(80, 200)cx, cy 0, 0pts self._rect(cx, cy, w, h)tool_r 3.0for seq, (x, y) in enumerate(pts):for strategy in [顺铣, 逆铣]:rows.append({contour_id: fCTR-{i1:02d},x: x, y: y, seq:利用AI解决实际问题如果你觉得这个工具好用欢迎关注长安牧笛