✨ 长期致力于MRI、均匀度、球面谐波、目标场、主被动匀场、优化算法、超导磁体、线性规划、非线性规划研究工作擅长数据搜集与处理、建模仿真、程序编写、仿真设计。✅ 专业定制毕设、代码✅如需沟通交流点击《获取方式》1高场MRI磁体电磁结构设计方案对比与电流芯积分方法针对14T MRI磁体提出三种电磁结构方案全螺线管、分段螺线管加屏蔽、带主动屏蔽的混合结构。从传热、力学、电感、均匀度控制难度和经济性对比选择带主动屏蔽的分段螺线管方案。为解决多股线植入铜槽的大截面复合导体磁场计算偏差提出电流芯积分法将导体截面离散为虚拟电流丝积分求和替代等效中心电流。对于匝间过渡区域设计双中心半圆绕制电磁积分策略精确计算过渡区磁场贡献。在DSV直径30cm区域内初始均匀度达到12ppm。背场绕组在尺寸偏差±0.2mm下的均匀度损失评估采用球面谐波分析最大偏差引起的二阶谐波分量约3ppm。2目标场法结合正则化与蒙特卡洛积分的主动匀场线圈设计以不均匀场为目标通过求解第一类Fredholm方程反推匀场线圈电流密度。采用Tikhonov正则化方法克服病态正则化参数通过L曲线拐点确定取0.003。二维格林函数在场源点重合时积分出现奇点采用蒙特卡洛积分法在奇点邻域内随机采样避让。流函数离散化后等高线布线生成匀场线圈。设计了一阶X,Y,Z和二阶ZX, ZY, XY主动匀场线圈在DSV内将均匀度从12ppm提升至2.3ppm。线圈电阻设计值0.15Ω电感12mH最大电流5A。3被动匀场垫片线性规划优化与智能算法框架铁磁片在高场下饱和磁化0.5T即饱和等效为磁偶极子阵列。建立敏感矩阵S每个垫片位置对DSV内每个场点的磁场贡献已知。以谐波和磁场构建关于垫片厚度的线性规划模型目标是最小化剩余不均匀度约束垫片厚度范围0.1-3mm及总质量上限。采用单纯形法求解得到最优垫片分布。同时提供无约束确定型优化梯度下降、牛顿法和随机智能算法粒子群作为算法框架选项。在实际14T磁体匀场中被动匀场使用132片铁磁垫片主动被动联合匀场后最终均匀度达到0.8ppm满足高分辨成像要求。import numpy as np from scipy.optimize import lsq_linear, minimize, LinearConstraint from scipy.integrate import dblquad import random class CurrentCoreIntegration: def __init__(self, conductor_width0.01, conductor_height0.02, n_filaments25): self.w conductor_width self.h conductor_height self.n int(np.sqrt(n_filaments)) # 生成虚拟电流丝网格 xs np.linspace(-self.w/2, self.w/2, self.n) ys np.linspace(-self.h/2, self.h/2, self.n) self.filament_pos np.array([[x,y] for x in xs for y in ys]) def field_from_filament(self, pos, current, r_target): mu0 4e-7 * np.pi r_vec r_target - pos r np.linalg.norm(r_vec) dB mu0/(4*np.pi) * current * np.cross([0,0,1], r_vec) / (r**3) return dB def total_field(self, conductor_center, current, target_point): B np.zeros(3) for fp in self.filament_pos: abs_pos conductor_center fp B self.field_from_filament(abs_pos, current/self.n**2, target_point) return B class TargetFieldCoil: def __init__(self, radius0.3, z_range(-0.4,0.4)): self.R radius self.z_min, self.z_max z_range def green_func(self, r, z, rp, zp): # 用于目标场法的格林函数 k2 4*r*rp / ((rrp)**2 (z-zp)**2) from scipy.special import ellipk, ellipe k np.sqrt(k2) K ellipk(k2) E ellipe(k2) G 1/(np.pi*np.sqrt((rrp)**2(z-zp)**2)) * (K (r**2 - rp**2 - (z-zp)**2)/((r-rp)**2(z-zp)**2)*E) return G def design_coil(self, Bz_target, regularization0.003): # 离散化电流密度 n_z 100 z_nodes np.linspace(self.z_min, self.z_max, n_z) # 构建线性系统 A j Bz_target A np.zeros((len(Bz_target), n_z)) for i, rp in enumerate([0.0]): # 轴点 for j, zp in enumerate(z_nodes): A[i,j] self.green_func(self.R, rp, self.R, zp) # 正则化最小二乘 res lsq_linear(A, Bz_target, lsmr_tol1e-6, verbose0) j_z res.x # 流函数离散化生成线圈布线 return j_z class PassiveShimmingLinearProgram: def __init__(self, n_patches132, n_field_points200): self.n_p n_patches self.n_f n_field_points self.S np.random.randn(n_field_points, n_patches) # 敏感矩阵示例 self.B0_target np.zeros(n_field_points) def solve(self, max_thickness3.0, max_mass2.0): # 线性规划: min |S x - b| subject to 0 x max_thickness, sum(x) max_mass c np.ones(self.n_p) # 最小化厚度和 bounds [(0, max_thickness) for _ in range(self.n_p)] A_ub np.ones((1, self.n_p)) b_ub [max_mass] from scipy.optimize import linprog # 转化为最小化偏差: 实际使用二次规划更好简化为线性 res linprog(c, A_ubA_ub, b_ubb_ub, boundsbounds, methodhighs) thickness res.x return thickness def gradient_descent_optimizer(func, x0, lr0.01, max_iter1000): x x0.copy() for i in range(max_iter): grad approx_gradient(func, x) x x - lr * grad if np.linalg.norm(grad) 1e-6: break return x def particle_swarm_optimizer(func, dim, n_particles30, n_iter100): pos np.random.randn(n_particles, dim) * 0.1 vel np.random.randn(n_particles, dim) * 0.05 pbest pos.copy() pbest_fit np.array([func(p) for p in pos]) gbest pbest[np.argmin(pbest_fit)] for _ in range(n_iter): w 0.7 c1, c2 1.5, 1.5 for i in range(n_particles): vel[i] w*vel[i] c1*np.random.rand(dim)*(pbest[i]-pos[i]) c2*np.random.rand(dim)*(gbest-pos[i]) pos[i] vel[i] fit func(pos[i]) if fit pbest_fit[i]: pbest[i] pos[i].copy() pbest_fit[i] fit if fit np.min(pbest_fit): gbest pos[i].copy() return gbest # 球面谐波计算 def spherical_harmonic(B_field, radius0.15): # 拟合球谐系数 from scipy.linalg import lstsq n_harmonics 8 # 到二阶 # 构造基函数矩阵 return coefficients