/// <summary>
/// 計算多變量管制圖的 Tsquare decomposition
/// </summary>
/// <param name="data"></param>
/// <param name="mean"></param>
/// <param name="S"></param>
/// <returns></returns>
public static LinearAlgebra.Matrix <double> T2Decomposition(LinearAlgebra.Matrix <double> data, LinearAlgebra.Matrix <double> mean, LinearAlgebra.Matrix <double> S)
{
//mean 的 #column = #columns of data = #columns of covariance
if (data.ColumnCount != mean.ColumnCount || data.ColumnCount != S.ColumnCount || mean.ColumnCount != S.ColumnCount)
{
throw new ArgumentException("矩陣的維度無法計算");
}
LinearAlgebra.Matrix <double> invS = S.Inverse();
List <double> tsquare = new List <double>();
for (int i = 0; i < data.RowCount; i++)
{
tsquare.Add(CalculateTSquare(data.Row(i).ToRowMatrix(), mean, invS));
}
List <double> decoValue = new List <double>();
for (int c = 0; c < data.ColumnCount; c++)
{
LinearAlgebra.Matrix <double> subData = data.RemoveColumn(c);
LinearAlgebra.Matrix <double> subMean = mean.RemoveColumn(c);
LinearAlgebra.Matrix <double> subS = S.RemoveColumn(c).RemoveRow(c);
LinearAlgebra.Matrix <double> invSubS = subS.Inverse();
for (int r = 0; r < data.RowCount; r++)
{
decoValue.Add(tsquare[r] - CalculateTSquare(subData.Row(r).ToRowMatrix(), subMean, invSubS));
}
}
LinearAlgebra.Matrix <double> m = LinearAlgebra.Matrix <double> .Build.DenseOfColumnMajor(data.RowCount, data.ColumnCount, decoValue);
return(m);
}
public LinearAlgebra.Vector <double> solveEquations(LinearAlgebra.Matrix <double> stiffnessMatrix, LinearAlgebra.Vector <double> forceVector, List <int> boundaryDofs, List <double> boundaryConstraints)
{
int nDof = forceVector.Count;
// Find all dofs where force is known
List <int> allDofs = Enumerable.Range(0, nDof).ToList();
List <int> unknownDofs = allDofs.Except(boundaryDofs).ToList();
// Add the know displacements to the result
LinearAlgebra.Vector <double> displacementVector = LinearAlgebra.Vector <double> .Build.Dense(nDof);
for (int i = 0; i < boundaryDofs.Count; i++)
{
displacementVector[boundaryDofs[i]] = boundaryConstraints[i];
}
// Pick out part of matrix corresponding to known forces
int nrUnknownDofs = unknownDofs.Count;
LinearAlgebra.Matrix <double> unknownK = LinearAlgebra.Matrix <double> .Build.Dense(nrUnknownDofs, nrUnknownDofs);
LinearAlgebra.Vector <double> knownForces = LinearAlgebra.Vector <double> .Build.Dense(unknownDofs.Count);
for (int i = 0; i < unknownDofs.Count; i++)
{
for (int j = 0; j < unknownDofs.Count; j++)
{
unknownK[i, j] = stiffnessMatrix[unknownDofs[i], unknownDofs[j]];
}
knownForces[i] = forceVector[unknownDofs[i]];
}
LinearAlgebra.Matrix <double> unkownKnownK = LinearAlgebra.Matrix <double> .Build.Dense(nrUnknownDofs, boundaryDofs.Count);
for (int i = 0; i < unknownDofs.Count; i++)
{
for (int j = 0; j < boundaryDofs.Count; j++)
{
unkownKnownK[i, j] = stiffnessMatrix[unknownDofs[i], boundaryDofs[j]];
}
knownForces[i] = forceVector[unknownDofs[i]];
}
// Solve for the unknown displacements
LinearAlgebra.Vector <double> unknownDisplacements = unknownK.Inverse().Multiply(knownForces.Subtract(unkownKnownK.Multiply(LinearAlgebra.Vector <double> .Build.Dense(boundaryConstraints.ToArray()))));
// Insert the calculated displacements
for (int i = 0; i < unknownDofs.Count; i++)
{
displacementVector[unknownDofs[i]] = unknownDisplacements[i];
}
return(displacementVector);
}
/// <summary>
/// 計算Tsquare
/// </summary>
/// <param name="item">a observation</param>
/// <param name="mean">mean vector</param>
/// <param name="invS">inverse of covariance matrix</param>
/// <returns></returns>
public static double CalculateTSquare(LinearAlgebra.Matrix <double> item, LinearAlgebra.Matrix <double> mean, LinearAlgebra.Matrix <double> invS)
{
if (item.ColumnCount != mean.ColumnCount || item.ColumnCount != invS.ColumnCount || mean.ColumnCount != invS.ColumnCount)
{
throw new ArgumentException("矩陣的維度無法計算");
}
LinearAlgebra.Matrix <double> diff = item - mean;
var t2 = diff.Multiply(invS).Multiply(diff.Transpose());
return(t2.At(0, 0));
}
/// <summary>
/// 計算數列各元素的累積和
/// </summary>
/// <param name="x">要處理的陣列，合法的輸入是 double[]</param>
/// <returns></returns>
public static double[] PartialSum(double[] x)
{
LinearAlgebra.Vector <double> vx
= LinearAlgebra.Double.DenseVector.OfArray(x);
int row = vx.Count;
int col = row;
LinearAlgebra.Matrix <double> lmat
= LinearAlgebra.Matrix <double> .Build.Dense(row, col, 1);
lmat = lmat.LowerTriangle();
LinearAlgebra.Vector <double> result
= lmat.Multiply(vx);
return(result.ToArray());
}
public LinearAlgebra.Matrix <double> ComputeStiffnessMatrix()
{
LinearAlgebra.Matrix <double> K = LinearAlgebra.Matrix <double> .Build.Dense(2, 2);
double x1 = Nodes[0].Point.X;
double y1 = Nodes[0].Point.Y;
double z1 = Nodes[0].Point.Z;
double x2 = Nodes[1].Point.X;
double y2 = Nodes[1].Point.Y;
double z2 = Nodes[1].Point.Z;
ElemLength = Math.Sqrt((x2 - x1) * (x2 - x1) + (y2 - y1) * (y2 - y1) + (z2 - z1) * (z2 - z1));
K[0, 0] = Area * StiffnessModulus / ElemLength;
K[0, 1] = -Area * StiffnessModulus / ElemLength;
K[1, 0] = -Area * StiffnessModulus / ElemLength;
K[1, 1] = Area * StiffnessModulus / ElemLength;
// Tranform to global element stiffness matrix using transformation matrix G
double nxx = (x2 - x1) / ElemLength;
double nyx = (y2 - y1) / ElemLength;
double nzx = (z2 - z1) / ElemLength;
G[0, 0] = nxx;
G[0, 1] = nyx;
G[0, 2] = nzx;
G[1, 3] = nxx;
G[1, 4] = nyx;
G[1, 5] = nzx;
LinearAlgebra.Matrix <double> GT = G.Transpose();
LinearAlgebra.Matrix <double> KGlobal = (GT.Multiply(K)).Multiply(G);
return(KGlobal);
}
protected override void SolveInstance(IGH_DataAccess DA)
{
// Retrive data from component
List <double> A = new List <double>();
List <double> E = new List <double>();
List <Line> lines = new List <Line>();
List <ContstraintNode> rNodes = new List <ContstraintNode>();
List <LoadNode> loadNodes = new List <LoadNode>();
double scaleFactor = 1.0;
DA.GetDataList("Line", lines);
DA.GetDataList("Area", A);
DA.GetDataList("Youngs Modulus", E);
DA.GetDataList("Restraint Nodes", rNodes);
DA.GetDataList("Load Nodes", loadNodes);
DA.GetData("Scale Factor", ref scaleFactor);
// The length of A and E must be the same as lines
if ((A.Count != E.Count) | (E.Count != lines.Count))
{
throw new ArgumentException("Length of A and E must equal length of Line");
}
// Create one list to store the nodes and one list to store the bars
List <Node> trussNodes = new List <Node>();
List <Bar> trussBars = new List <Bar>();
// Topology matrix to keep track of element dofs
List <List <int> > eDof = new List <List <int> >();
// Loop trough each line and create nodes at end points
for (int i = 0; i < lines.Count; i++)
{
Node node1 = new Node(lines[i].From);
Node node2 = new Node(lines[i].To);
// To keep track if the node is unique
bool unique1 = true;
bool unique2 = true;
// Check if node is unique, if so give it an ID and degress of freedom
foreach (Node existingNode in trussNodes)
{
// If not unique use an already identified node
if (node1 == existingNode)
{
node1 = existingNode;
unique1 = false;
}
if (node2 == existingNode)
{
node2 = existingNode;
unique2 = false;
}
}
// If unique give it an ID
if (unique1)
{
int id_node_1 = trussNodes.Count;
node1.ID = id_node_1;
node1.Dofs = System.Linq.Enumerable.Range(id_node_1 * 3, 3).ToList();
// Check if any boundary node or load node exist at current node
foreach (ContstraintNode rNode in rNodes)
{
if (rNode == node1)
{
// Add restraint data
node1.ConstraintX = rNode.ConstraintX;
node1.ConstraintY = rNode.ConstraintY;
node1.ConstraintZ = rNode.ConstraintZ;
}
}
foreach (LoadNode loadNode in loadNodes)
{
if (loadNode == node1)
{
// Add force data
node1.ForceX = loadNode.ForceX;
node1.ForceY = loadNode.ForceY;
node1.ForceZ = loadNode.ForceZ;
}
}
// Finally add the node
trussNodes.Add(node1);
}
if (unique2)
{
int id_node_2 = trussNodes.Count;
node2.ID = id_node_2;
node2.Dofs = System.Linq.Enumerable.Range(id_node_2 * 3, 3).ToList();
// Check if any boundary node or load node exist at current node
foreach (ContstraintNode rNode in rNodes)
{
if (rNode == node2)
{
// Add constraint data
node2.ConstraintX = rNode.ConstraintX;
node2.ConstraintY = rNode.ConstraintY;
node2.ConstraintZ = rNode.ConstraintZ;
}
}
foreach (LoadNode loadNode in loadNodes)
{
if (loadNode == node2)
{
// Add force data
node2.ForceX = loadNode.ForceX;
node2.ForceY = loadNode.ForceY;
node2.ForceZ = loadNode.ForceZ;
}
}
// Finally add the node
trussNodes.Add(node2);
}
// Create a bar object between the nodes
Bar bar = new Bar(node1, node2, A[i], E[i]);
trussBars.Add(bar);
// Topology matrix
List <int> dofs1 = bar.Nodes[0].Dofs;
List <int> dofs2 = bar.Nodes[1].Dofs;
List <int> eDofRow = new List <int>();
eDofRow.AddRange(dofs1);
eDofRow.AddRange(dofs2);
eDof.Add(eDofRow);
}
int nDof = trussNodes.Count * 3;
int nElem = eDof.Count;
// Loop trough each node and construct a load vector and boundary vector
LinearAlgebra.Vector <double> forceVector = LinearAlgebra.Vector <double> .Build.Dense(nDof);
List <int> boundaryDofs = new List <int>();
List <double?> boundaryConstraints = new List <double?>();
for (int i = 0; i < trussNodes.Count; i++)
{
// Load vector
forceVector[i * 3] = trussNodes[i].ForceX;
forceVector[i * 3 + 1] = trussNodes[i].ForceY;
forceVector[i * 3 + 2] = trussNodes[i].ForceZ;
// Boundary vector
for (int j = 0; j < 3; j++)
{
if (j == 0 && trussNodes[i].ConstraintX != null)
{
boundaryDofs.Add(trussNodes[i].Dofs[j]);
boundaryConstraints.Add(trussNodes[i].ConstraintX);
}
else if (j == 1 && trussNodes[i].ConstraintY != null)
{
boundaryDofs.Add(trussNodes[i].Dofs[j]);
boundaryConstraints.Add(trussNodes[i].ConstraintY);
}
else if (j == 2 && trussNodes[i].ConstraintZ != null)
{
boundaryDofs.Add(trussNodes[i].Dofs[j]);
boundaryConstraints.Add(trussNodes[i].ConstraintZ);
}
}
}
// Loop trough each element, compute local stiffness matrix and assemble into global stiffness matrix
LinearAlgebra.Matrix <double> K = LinearAlgebra.Matrix <double> .Build.Dense(nDof, nDof);
for (int i = 0; i < trussBars.Count; i++)
{
LinearAlgebra.Matrix <double> KElem = trussBars[i].ComputeStiffnessMatrix();
// Assemble
for (int k = 0; k < 6; k++)
{
for (int l = 0; l < 6; l++)
{
K[eDof[i][k], eDof[i][l]] = K[eDof[i][k], eDof[i][l]] + KElem[k, l];
}
}
}
// Calculate the displacements
Solver solver = new Solver();
LinearAlgebra.Vector <double> displacements = solver.solveEquations(K, forceVector, boundaryDofs, boundaryConstraints.Cast <double>().ToList());
// Save the displacement for each node and calculate the stress in each bar
List <double> elemStress = new List <double> {
};
for (int i = 0; i < nElem; i++)
{
double disp1 = displacements[eDof[i][0]];
double disp2 = displacements[eDof[i][1]];
double disp3 = displacements[eDof[i][2]];
double disp4 = displacements[eDof[i][3]];
double disp5 = displacements[eDof[i][4]];
double disp6 = displacements[eDof[i][5]];
Point3d newPoint1 = new Point3d(disp1 * scaleFactor, disp2 * scaleFactor, disp3 * scaleFactor);
Point3d newPoint2 = new Point3d(disp4 * scaleFactor, disp5 * scaleFactor, disp6 * scaleFactor);
// Translate original points
trussBars[i].Nodes[0].Point = trussBars[i].Nodes[0].Point + newPoint1;
trussBars[i].Nodes[1].Point = trussBars[i].Nodes[1].Point + newPoint2;
// Calculate element stress
elemStress.Add(trussBars[i].ComputeStress(new List <double> {
disp1, disp2, disp3, disp4, disp5, disp6
}));
}
// Return the deformed lines
List <Line> deformedLines = new List <Line>();
foreach (Bar bar in trussBars)
{
deformedLines.Add(new Line(bar.Nodes[0].Point, bar.Nodes[1].Point));
}
DA.SetDataList("Deformed Truss", deformedLines);
DA.SetDataList("Element Stress", elemStress);
}
public override void Execute(Mtb.Project proj)
{
// 將資料匯入 Minitab
if (_rawdata == null || _rawdata.Rows.Count == 0)
{
return;
}
_rptLst = new List <IRptOutput>(); //重新建立一個分析結果列舉
Mtb.Worksheet ws = proj.Worksheets.Add(1); //新增工作表
Mtblib.Tools.MtbTools.InsertDataTableToMtbWs(_rawdata, ws); //匯入資料至 Minitab
List <string> varnames = new List <string>();
List <Mtb.Column> varCols = new List <Mtb.Column>(); //變數的欄位集合
for (int i = 0; i < _rawdata.Columns.Count; i++)
{
DataColumn col = _rawdata.Columns[i];
switch (col.ColumnName)
{
case "TIMESTAMP":
case "CHART_PARA_INDEX":
break;
default:
varnames.Add(col.ColumnName);
varCols.Add(ws.Columns.Item(col.ColumnName));
break;
}
}
foreach (var col in varCols)
{
if ((int)col.DataType == 3 || col.MissingCount == col.RowCount)
{
throw new ArgumentNullException(string.Format("[{0}]查無資料-多變量管制圖", col.Name));
}
}
StringBuilder cmnd = new StringBuilder();
List <TsquareParameters> tmpParaList = new List <TsquareParameters>(); //用於計算 decomposition
string[] colIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, 4, Mtblib.Tools.MtbVarType.Column); //Plot point, CL, UCL and Test column
// 指定繪圖時需要的欄位變數
Mtb.Column
pplotCol = ws.Columns.Item(colIds[0]),
clCol = ws.Columns.Item(colIds[1]),
uclCol = ws.Columns.Item(colIds[2]),
oocCol = ws.Columns.Item(colIds[3]),
timeCol = ws.Columns.Item("TIMESTAMP");
// 計算 T2 plot points
if (_parameters == null || _parameters.Count == 0)
{
//自己算的時候需要的變數和子命令
#region Phase I
cmnd.AppendLine("macro");
cmnd.AppendLine("myt2calculator x.1-x.p;");
cmnd.AppendLine("variance cov ssiz;");
cmnd.AppendLine("meanvect location; ");
cmnd.AppendLine("ppoint t2;");
cmnd.AppendLine("climit cl ucl;");
cmnd.AppendLine("test ooc;");
cmnd.AppendLine("siglevel alpha.");
cmnd.AppendLine("mcolumn x.1-x.p t2");
cmnd.AppendLine("mcolumn loc.1-loc.p xx.1-xx.p tmp.1-tmp.3 ooc ucl cl");
cmnd.AppendLine("mconstant ssiz m alpha conf a1 a2 kk");
cmnd.AppendLine("mmatrix location diff cov tdiff invCov");
cmnd.AppendLine("default alpha = 0.0027");
cmnd.AppendLine("rnmiss x.1-x.p tmp.1");
cmnd.AppendLine("copy x.1-x.p xx.1-xx.p;");
cmnd.AppendLine("exclude;");
cmnd.AppendFormat("where \"tmp.1>0\".\r\n");
cmnd.AppendLine("cova xx.1-xx.p cov"); // Get Covariance matrix
cmnd.AppendLine("let ssiz = n(xx.1)"); // Get the sample size of covariance calculation, ALSO..THAT IS THE NUMBER OF NONMISSING OBSERVATIONS
cmnd.AppendLine("stat x.1-x.p;");
cmnd.AppendLine("mean loc.1-loc.p."); //Get mean vector.
/*
* Check if there is enough nonmissing observation to calculate control limit under
* Tsquare command (m>p+1, where m=#observation, include missing obs, p=#items)
* If no, you still need to draw plot points on graph..
* The trick is we given parameters (Mean & Covariance) and given a fake sample size, then
* you get t-sq value, becasue we don't need the CL & UCL from Minitab.
* (WE USE REGULAR COVARIANCE INSTEAD OF COVARIANCE BY Sullivan & Woodall)
*
*/
cmnd.AppendLine("if(ssiz < p+1)");
cmnd.AppendLine("let kk = p+1");
cmnd.AppendLine("else");
cmnd.AppendLine("copy ssiz kk\r\n");
cmnd.AppendLine("endif");
cmnd.AppendLine("tsquare x.1-x.p 1;");
cmnd.AppendLine("mu loc.1-loc.p;");
cmnd.AppendLine("sigma cov;");
cmnd.AppendLine("number kk;");
cmnd.AppendLine("sampsize tmp.1;"); //Get subgroup size.
cmnd.AppendLine("ppoint t2."); //Get Tsquare value.
cmnd.AppendLine("copy loc.1-loc.p location"); //Copy column to mean vector
cmnd.AppendLine("let m = sum(tmp.1)"); //Get the actual sample size (include missing observations)
cmnd.AppendLine("if (m <= p+1)"); // Check if there is enough obs to calc contril limit...
cmnd.AppendLine("let cl[m]=miss()"); //no center line
cmnd.AppendLine("let ucl[m]=miss()"); //no control limit
cmnd.AppendLine("set ooc"); //no ooc
cmnd.AppendLine("(0)m");
cmnd.AppendLine("end");
cmnd.AppendLine("else");
cmnd.AppendLine("let conf = 1-alpha");
cmnd.AppendLine("let a1 = p/2");
cmnd.AppendLine("let a2 = (m-p-1)/2");
// Calculate UCL
cmnd.AppendLine("set tmp.1");
cmnd.AppendLine("(conf)m");
cmnd.AppendLine("end");
cmnd.AppendLine("invcdf tmp.1 ucl;");
cmnd.AppendLine(" beta a1 a2.");
cmnd.AppendLine("let ucl = (m-1)**2/m*ucl"); //Get upper control limit
// Calculate CL
cmnd.AppendLine("set tmp.1");
cmnd.AppendLine("(0.5)m");
cmnd.AppendLine("end");
cmnd.AppendLine("invcdf tmp.1 cl;");
cmnd.AppendLine(" beta a1 a2.");
cmnd.AppendLine("let cl = (m-1)**2/m*cl"); //Get center line
cmnd.AppendLine("let ooc = if(t2>ucl and t2<>MISS(),1,0)"); //Get OOC info
cmnd.AppendLine("endif");
//刪除所有圖形
cmnd.AppendLine("gmana;");
cmnd.AppendLine("all;");
cmnd.AppendLine("close;");
cmnd.AppendLine("nopr.");
cmnd.AppendLine("endmacro");
string macPath = Mtblib.Tools.MtbTools.BuildTemporaryMacro("mytsquare.mac", cmnd.ToString());
string[] matIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, 2, Mtblib.Tools.MtbVarType.Matrix); //紀錄Mean vecot & Covariance matrix
string[] constIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, 1, Mtblib.Tools.MtbVarType.Constant); // Sample size of Covariance matrix
cmnd.Clear();
cmnd.AppendLine("notitle");
cmnd.AppendLine("brief 0");
cmnd.AppendFormat("%\"{0}\" {1};\r\n", macPath, string.Join(" &\r\n", varCols.Select(x => x.SynthesizedName)));
cmnd.AppendFormat("variance {0} {1};\r\n", matIds[0], constIds[0]);
cmnd.AppendFormat("meanvect {0};\r\n", matIds[1]);
cmnd.AppendFormat("ppoint {0};\r\n", pplotCol.SynthesizedName);
cmnd.AppendFormat("climit {0} {1};\r\n", clCol.SynthesizedName, uclCol.SynthesizedName);
cmnd.AppendFormat("test {0}.\r\n", oocCol.SynthesizedName);
string path = Mtblib.Tools.MtbTools.BuildTemporaryMacro("mymacro.mtb", cmnd.ToString());
proj.ExecuteCommand(string.Format("exec \"{0}\" 1", path), ws);
#endregion
//取得參數組
TsquareParameters tmpPara = new TsquareParameters();
Mtb.Matrix mat;
mat = ws.Matrices.Item(matIds[1]);
tmpPara.Mean = LinearAlgebra.Matrix <double> .Build.DenseOfColumnMajor(1, mat.ColumnCount, mat.GetData());
mat = ws.Matrices.Item(matIds[0]);
tmpPara.Covariance = LinearAlgebra.Matrix <double> .Build.DenseOfColumnMajor(mat.RowCount, mat.ColumnCount, mat.GetData());
tmpPara.SampleSize = ws.Constants.Item(constIds[0]).GetData();
tmpPara.SubgroupSize = 1;
tmpPara.ParaID = null;
tmpParaList.Add(tmpPara);
}
else
{
//有指定參數的時候
#region Phase II
int nParas = ws.Columns.Item("CHART_PARA_INDEX").GetNumDistinctRows(); //宣告多組參數組
cmnd.AppendLine("macro");
cmnd.AppendLine("myt2calculator x.1-x.p;");
cmnd.AppendFormat("variance cov.1-cov.{0} ssiz.1-ssiz.{0};\r\n", nParas);
cmnd.AppendFormat("meanvect mvect.1-mvect.{0};\r\n", nParas);
cmnd.AppendLine("ppoint t2;");
cmnd.AppendLine("climit cl ucl;");
cmnd.AppendLine("test ooc;");
cmnd.AppendLine("siglevel alpha.");
cmnd.AppendLine("mcolumn x.1-x.p xx.1-xx.p xxx.1-xxx.p loc.1-loc.p");
cmnd.AppendLine("mcolumn tmp t2 ucl cl ooc tmpT2 tmpUCL tmpCL tmpOOC");
cmnd.AppendFormat("mmatrix mvect.1-mvect.{0} cov.1-cov.{0}\r\n", nParas);
cmnd.AppendFormat("mconstant m histm alpha conf a1 a2 ssiz.1-ssiz.{0} kk\r\n", nParas);
cmnd.AppendLine("default alpha = 0.0027");
var _chartparaindex = _rawdata.AsEnumerable()
.Select(x => x.Field <string>("CHART_PARA_INDEX")).ToArray();
var _distinctIds = _chartparaindex.Distinct().ToArray();
for (int i = 0; i < _distinctIds.Length; i++) //對每個參數與對應的資料計算 Tsquare 和管制界限
{
string id = _distinctIds[i];
cmnd.AppendLine("copy x.1-x.p xx.1-xx.p;");
cmnd.AppendLine("include;");
cmnd.AppendFormat("rows {0}.\r\n",
string.Join(" &\r\n",
_chartparaindex.Select((x, rowid) => new { Value = x, RowId = rowid + 1 })
.Where(x => x.Value == id).Select(x => x.RowId).ToArray()));
#region Set parameter information into Minitab macro
TsquareParameters para = new TsquareParameters();
para.Mean = null; para.Covariance = null; para.SampleSize = MISSINGVALUE; para.SubgroupSize = 1;
if (id != null)
{
para = _parameters.Where(x => x.ParaID == id).First();
}
if (para.Mean != null) //把已知的 Mean vector 寫入 Minitab
{
cmnd.AppendLine("read loc.1-loc.p");
cmnd.AppendFormat("{0}\r\n", string.Join(" &\r\n", para.Mean.Enumerate()));
cmnd.AppendLine("end");
}
else
{
cmnd.AppendLine("stat xx.1-xx.p;");
cmnd.AppendLine("mean loc.1-loc.p.");
}
if (para.Covariance != null) //把已知的 Covariance 寫入 Minitab
{
cmnd.AppendFormat("read {1} {2} cov.{0} \r\n", i + 1, para.Covariance.RowCount, para.Covariance.ColumnCount);
List <LinearAlgebra.Vector <double> > valuesByRow = para.Covariance.EnumerateRows().ToList();
for (int r = 0; r < para.Covariance.RowCount; r++)
{
cmnd.AppendFormat("{0}\r\n", string.Join(" &\r\n", valuesByRow[r]));
}
cmnd.AppendLine("end");
cmnd.AppendFormat("let ssiz.{0}={1}\r\n", i + 1, para.SampleSize);
}
else
{
cmnd.AppendLine("rnmiss xx.1-xx.p tmp");
cmnd.AppendLine("copy xx.1-xx.p xxx.1-xxx.p;");
cmnd.AppendLine("exclud;");
cmnd.AppendFormat("where \"tmp>0\".\r\n");
cmnd.AppendFormat("cova xxx.1-xxx.p cov.{0}\r\n", i + 1);
cmnd.AppendFormat("let ssiz.{0} = count(xxx.1)\r\n", i + 1);
}
#endregion
//Tsquare command
cmnd.AppendFormat("if(ssiz.{0} < p+1)\r\n", i + 1);
cmnd.AppendLine("let kk = p+1");
cmnd.AppendLine("else");
cmnd.AppendFormat("copy ssiz.{0} kk\r\n", i + 1);
cmnd.AppendLine("endif");
if (id == null && _chartparaindex.Where(x => x == id).Count() == 1) //如果只有一個觀測值且沒有參數設定
{
cmnd.AppendLine("let tmp=1");
cmnd.AppendLine("let tmpT2=miss()");
}
else
{
cmnd.AppendFormat("tsquare xx.1-xx.p {0};\r\n", para.SubgroupSize);
cmnd.AppendLine(" mu loc.1-loc.p;");
cmnd.AppendFormat(" sigma cov.{0};\r\n", i + 1);
cmnd.AppendLine(" number kk;");
cmnd.AppendLine("sampsize tmp;"); //Get subgroup size.
cmnd.AppendLine("ppoint tmpT2."); //Get Tsquare value.
}
cmnd.AppendFormat("copy loc.1-loc.p mvect.{0}\r\n", i + 1); //Copy column to mean vector
/*
* Get the value of m which use to calculate UCL, CL.
* In phase II, m is the sample size of the data used to calculate historical
* covariance matrix.
* In phase I, m is the number of observations.
*
*/
cmnd.AppendLine("let m = sum(tmp)"); //Get the actual sample size (include missing data)
if (para.Covariance != null)
{
cmnd.AppendFormat("let histm = ssiz.{0}\r\n", i + 1);
}
else
{
cmnd.AppendLine("copy m histm");
}
// Calculate control limits
if (id == null && _chartparaindex.Where(x => x == id).Count() == 1)
{
cmnd.AppendLine("let tmpUCL=miss()");
cmnd.AppendLine("let tmpCL=miss()");
}
else
{
cmnd.AppendLine("let conf = 1-alpha");
cmnd.AppendLine("let a1 = p");
cmnd.AppendLine("let a2 = histm");
#region Get UCL
cmnd.AppendLine("set tmp");
cmnd.AppendLine("(conf)m");
cmnd.AppendLine("end");
if (id == null) //phase I case
{
cmnd.AppendLine("invcdf tmp tmpUCL;");
cmnd.AppendLine(" beta a1 a2.");
cmnd.AppendLine("let tmpUCL = (m-1)**2/m*tmpUCL"); //Get upper control limit
}
else //phase II case
{
cmnd.AppendLine("invcdf tmp tmpUCL;");
cmnd.AppendLine(" f a1 a2.");
cmnd.AppendLine("let tmpUCL = p*(histm+1)*(histm-1)/histm/(histm-p)*tmpUCL"); //Get upper control limit
}
#endregion
#region Get CL
cmnd.AppendLine("set tmp");
cmnd.AppendLine("(0.5)m");
cmnd.AppendLine("end");
if (id == null)
{
cmnd.AppendLine("invcdf tmp tmpCL;");
cmnd.AppendLine(" beta a1 a2.");
cmnd.AppendLine("let tmpCL = (m-1)**2/m*tmpCL"); //Get upper control limit
}
else
{
cmnd.AppendLine("invcdf tmp tmpCL;");
cmnd.AppendLine(" f a1 a2.");
cmnd.AppendLine("let tmpCL = p*(histm+1)*(histm-1)/histm/(histm-p)*tmpCL"); //Get center line
}
#endregion
}
cmnd.AppendLine("let tmpOOC = if(tmpT2>tmpUCL AND tmpT2<>MISS(),1,0)"); //Get OOC info
if (i == 0)
{
cmnd.AppendLine("copy tmpT2 tmpUCL tmpCL tmpOOC t2 ucl cl ooc");
}
else
{
cmnd.AppendLine("stack (t2 ucl cl ooc) (tmpT2 tmpUCL tmpCL tmpOOC) (t2 ucl cl ooc)");
}
}
//刪除所有圖形
cmnd.AppendLine("gmana;");
cmnd.AppendLine("all;");
cmnd.AppendLine("close;");
cmnd.AppendLine("nopr.");
cmnd.AppendLine("endmacro");
string macPath = Mtblib.Tools.MtbTools.BuildTemporaryMacro("mytsquare.mac", cmnd.ToString());
string[] matMeanIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, nParas, Mtblib.Tools.MtbVarType.Matrix);
string[] matCovIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, nParas, Mtblib.Tools.MtbVarType.Matrix);
string[] constIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, 1 * nParas, Mtblib.Tools.MtbVarType.Constant);
cmnd.Clear();
cmnd.AppendLine("notitle");
cmnd.AppendLine("brief 0");
cmnd.AppendFormat("%\"{0}\" {1};\r\n", macPath, string.Join(" &\r\n", varCols.Select(x => x.SynthesizedName)));
cmnd.AppendFormat("variance {0} {1};\r\n", string.Join(" &\r\n", matCovIds), string.Join(" &\r\n", constIds));
cmnd.AppendFormat("meanvect {0};\r\n", string.Join(" &\r\n", matMeanIds));
cmnd.AppendFormat("ppoint {0};\r\n", pplotCol.SynthesizedName);
cmnd.AppendFormat("climit {0} {1};\r\n", clCol.SynthesizedName, uclCol.SynthesizedName);
cmnd.AppendFormat("test {0}.\r\n", oocCol.SynthesizedName);
string path = Mtblib.Tools.MtbTools.BuildTemporaryMacro("mymacro.mtb", cmnd.ToString());
proj.ExecuteCommand(string.Format("exec \"{0}\" 1", path), ws);
#endregion
//取得參數組
TsquareParameters tmpPara;
Mtb.Matrix mat;
for (int i = 0; i < _distinctIds.Length; i++)
{
tmpPara = new TsquareParameters();
tmpPara.ParaID = _distinctIds[i];
mat = ws.Matrices.Item(matMeanIds[i]);
tmpPara.Mean = LinearAlgebra.Matrix <double> .Build.DenseOfColumnMajor(mat.RowCount, mat.ColumnCount, mat.GetData());
mat = ws.Matrices.Item(matCovIds[i]);
tmpPara.Covariance = LinearAlgebra.Matrix <double> .Build.DenseOfColumnMajor(mat.RowCount, mat.ColumnCount, mat.GetData());
tmpPara.SampleSize = ws.Constants.Item(constIds[i]).GetData();
tmpParaList.Add(tmpPara);
}
}
//繪圖
//會使用 TIMESTAMP, PPOINT, CL, UCL, TEST
#region 繪製 T2 Control Chart (TSPLOT)
string gpath = System.IO.Path.Combine(Environment.GetEnvironmentVariable("tmp"), "Minitab", string.Format("tsquare_{0}.jpg", _rawdata.TableName));
cmnd.Clear();
cmnd.AppendFormat("fdate {0};\r\n", timeCol.SynthesizedName);
cmnd.AppendLine("format(dtyyyy-MM-dd hh:mm).");
cmnd.AppendFormat("tsplot {0} {1} {2};\r\n", pplotCol.SynthesizedName, uclCol.SynthesizedName, clCol.SynthesizedName);
cmnd.AppendFormat("gsave \"{0}\";\r\n", gpath);
cmnd.AppendLine("repl;");
cmnd.AppendLine("jpeg;");
cmnd.AppendLine("over;");
cmnd.AppendFormat("symb {0};\r\n", oocCol.SynthesizedName);
cmnd.AppendLine("type &");
double[] oocData = oocCol.GetData();
if (oocData.Any(x => x == 0))
{
cmnd.AppendLine("6 &");
}
if (oocData.Any(x => x == 1))
{
cmnd.AppendLine("12 &");
}
cmnd.AppendLine("0 0 0 0;");
cmnd.AppendLine("size 1;");
cmnd.AppendLine("color &");
if (oocData.Any(x => x == 0))
{
cmnd.AppendLine("1 &"); //r17 color 64
}
if (oocData.Any(x => x == 1))
{
cmnd.AppendLine("2 &");
}
cmnd.AppendLine(";");
cmnd.AppendLine("conn;");
cmnd.AppendLine("type 1 1 1;");
cmnd.AppendLine("color 1 2 120;"); //r17 conn:64 cl: 9, climit:8
//cmnd.AppendLine("graph;");
//cmnd.AppendLine("color 22;");
cmnd.AppendLine("nole;");
cmnd.AppendFormat("stamp {0};\r\n", timeCol.SynthesizedName);
cmnd.AppendLine("scale 1;");
cmnd.AppendFormat("tick 1:{0}/{1};\r\n", pplotCol.RowCount,
pplotCol.RowCount > 35 ? Math.Ceiling((double)pplotCol.RowCount / 35) : 1);
cmnd.AppendLine("axla 1;");
cmnd.AppendLine("adis 0;");
cmnd.AppendLine("axla 2;");
cmnd.AppendLine("adis 0;");
string ttlString = string.Join(",", varnames);
if (ttlString.Length > 23)
{
ttlString = ttlString.Substring(0, 20) + "...";
}
cmnd.AppendLine("graph 8 4;");
cmnd.AppendFormat("title \"T2管制圖 {0}\";\r\n", ttlString);
cmnd.AppendFormat("footn \"更新時間: {0}\";\r\n", DateTime.Now);
cmnd.AppendFormat("ZTag \"{0}\";\r\n", "_T2CHART");
cmnd.AppendLine(".");
//刪除所有圖形
cmnd.AppendLine("gmana;");
cmnd.AppendLine("all;");
cmnd.AppendLine("close;");
cmnd.AppendLine("nopr.");
string t2MacroPath = Mtblib.Tools.MtbTools.BuildTemporaryMacro("myt2macro.mtb", cmnd.ToString());
proj.ExecuteCommand(string.Format("exec \"{0}\" 1", t2MacroPath), ws);
#endregion
//將檔案轉為二進位陣列
this.Contents.Add(new RptOutput()
{
OType = MtbOType.GRAPH,
OutputInByteArr = File.ReadAllBytes(gpath)
});
//計算 Decomposition
if (oocData.Any(x => x == 1))
{
DataTable tmpDataTable = _rawdata.Copy();
tmpDataTable.Columns.Add("OOC", typeof(int));
for (int r = 0; r < tmpDataTable.Rows.Count; r++)
{
DataRow dr = tmpDataTable.Rows[r];
dr["OOC"] = oocData[r];
}
//建立 Decomposition 的表格
DataTable decoTable = new DataTable();
decoTable.Columns.Add("TIMESTAMP", typeof(DateTime));
foreach (var item in varnames)
{
decoTable.Columns.Add(item, typeof(double));
}
//將OOC的項目值取出
var subData = tmpDataTable.Select("OOC=1").CopyToDataTable();
var oocParaSet = subData.AsEnumerable().Select(dr => dr.Field <string>("CHART_PARA_INDEX")).Distinct().ToArray(); //OOC 的參數組有哪些
foreach (var item in tmpParaList)
{
if (oocParaSet.Contains(item.ParaID)) //該參數組的觀測值是OOC
{
//把對應的觀測值轉換成 List<double[]>，其中每個 double[] 是每個 row 的數據
var subsubData = subData.Select(string.Format("CHART_PARA_INDEX {0}", item.ParaID == null ? "IS NULL" : "='" + item.ParaID + "'"))
.CopyToDataTable();
var obsArray = subsubData.DefaultView.ToTable(false, varnames.ToArray())
.AsEnumerable().Select(x => x.ItemArray.Select(o => Convert.ToDouble(o)).ToArray()).ToList();
LinearAlgebra.Matrix <double> obs = LinearAlgebra.Matrix <double> .Build.DenseOfRowArrays(obsArray);
LinearAlgebra.Matrix <double> t2deco = Tool.T2Decomposition(obs, item.Mean, item.Covariance);
var t2decoByRow = t2deco.EnumerateRows().ToArray();
for (int r = 0; r < t2decoByRow.Count(); r++)
{
DataRow dr = decoTable.NewRow();
object[] o = new object[1 + t2deco.ColumnCount];
t2decoByRow[r].ToArray().CopyTo(o, 1);
o[0] = subsubData.Rows[r].Field <DateTime>("TIMESTAMP");
decoTable.Rows.Add(o);
}
}
}
cmnd.Clear();
//LinearAlgebra.Matrix<double> obs = LinearAlgebra.Matrix<double>.Build.DenseOfRowArrays(subData);
this.Contents.Add(new RptOutput()
{
OType = MtbOType.TABLE,
OutputInByteArr = Tool.ConvertDataSetToByteArray(decoTable)
});
}
else // 沒有 OOC 就不做
{
this.Contents.Add(new RptOutput()
{
OType = MtbOType.TABLE,
OutputInByteArr = null
});
}
//Console.ReadKey();
}
public override void Execute(Mtb.Project proj)
{
// 將資料匯入 Minitab
if (_rawdata == null || _rawdata.Rows.Count == 0)
{
throw new ArgumentNullException("查無對應資料");
}
_rptLst = new List <IRptOutput>(); //重新建立一個分析結果列舉
proj.Commands.Delete();
proj.Worksheets.Delete();
Mtb.Worksheet ws = proj.Worksheets.Add(1); //新增工作表
Mtblib.Tools.MtbTools.InsertDataTableToMtbWs(_rawdata, ws); //匯入資料至 Minitab
List <Mtb.Column> varCols = new List <Mtb.Column>(); //變數的欄位集合
List <Mtb.Column> bkCols = new List <Mtb.Column>(); //斷絲率的集合
Mtb.Column timeCol = null;
for (int i = 0; i < _rawdata.Columns.Count; i++)
{
DataColumn col = _rawdata.Columns[i];
switch (col.ColumnName)
{
case "TIMESTAMP":
timeCol = ws.Columns.Item(col.ColumnName);
break;
case "RPT_TIMEHOUR":
case "GROUP_ID":
case "SITE_ID":
break;
default:
if (col.ColumnName.Substring(0, 2) == "BK")
{
bkCols.Add(ws.Columns.Item(col.ColumnName));
}
else
{
varCols.Add(ws.Columns.Item(col.ColumnName));
}
break;
}
}
foreach (var col in bkCols)
{
if ((int)col.DataType == 3 || col.MissingCount == col.RowCount)
{
throw new ArgumentNullException(string.Format("[{0}]查無資料", col.Name));
}
}
foreach (var col in varCols)
{
if ((int)col.DataType == 3 || col.MissingCount == col.RowCount)
{
throw new ArgumentNullException(string.Format("[{0}]查無資料", col.Name));
}
}
StringBuilder cmnd = new StringBuilder();
/*
* Create TSPlot
* 該圖會將數據於多個平面中顯示(k*1的方式)，所以要先設定 Panel 的安排方式
*
*/
cmnd.AppendLine("notitle");
cmnd.AppendLine("brief 0");
cmnd.AppendFormat("fdate {0};\r\n", timeCol.SynthesizedName);
cmnd.AppendLine("format(dtyyyy-MM-dd hh:mm).");
cmnd.AppendFormat("tsplot {0} {1};\r\n", string.Join(" &\r\n", bkCols.Select(x => x.SynthesizedName)),
string.Join(" &\r\n", varCols.Select(x => x.SynthesizedName)));
double gHeight = 3;
//if (varCols.Count > 3) gHeight = Math.Min(30, 4 + ((double)varCols.Count - 3) * 0.8);
cmnd.AppendFormat("graph 8 {0};\r\n", gHeight);
string gpath_trend = System.IO.Path.Combine(Environment.GetEnvironmentVariable("tmp"), "Minitab", string.Format("trend_{0}.jpg", _rawdata.TableName));
cmnd.AppendFormat("gsave \"{0}\";\r\n", gpath_trend);
cmnd.AppendLine("repl;");
cmnd.AppendLine("jpeg;");
//cmnd.AppendLine("panel;");
//cmnd.AppendFormat("rc {0} 1;\r\n", varCols.Count + bkCols.Count);
//cmnd.AppendLine("noal;");
//cmnd.AppendLine("label;");
//cmnd.AppendLine(" psize 5;");
cmnd.AppendLine("scale 1;");
cmnd.AppendLine(" psize 8;");
cmnd.AppendLine("grid 1;");
cmnd.AppendLine("axlab 1;");
cmnd.AppendLine("lshow;");
cmnd.AppendLine(" angle 90;");
cmnd.AppendLine("grid 2;");
cmnd.AppendLine("scale 2;");
cmnd.AppendLine(" psize 8;");
if (timeCol != null && (int)timeCol.DataType != 3)
{
cmnd.AppendFormat("stamp {0};\r\n", timeCol.SynthesizedName);
}
cmnd.AppendLine("symb;");
cmnd.AppendLine("conn;");
//cmnd.AppendFormat("foot \"建立時間: {0}\";\r\n", DateTime.Now);
cmnd.AppendLine("nodt.");
//Create Matrix Plot
cmnd.AppendFormat("matrixplot ({0})*({1});\r\n", string.Join(" &\r\n", bkCols.Select(x => x.SynthesizedName)),
string.Join(" &\r\n", varCols.Select(x => x.SynthesizedName)));
double gWidth = 8;
if (varCols.Count > 5)
{
gWidth = Math.Min(30, 8 + ((double)varCols.Count - 5) * 0.8);
}
cmnd.AppendFormat("graph {0} 5;\r\n", gWidth);
string gpath_corr = System.IO.Path.Combine(Environment.GetEnvironmentVariable("tmp"), "Minitab", string.Format("corr_{0}.jpg", _rawdata.TableName));
cmnd.AppendFormat("gsave \"{0}\";\r\n", gpath_corr);
cmnd.AppendLine("repl;");
cmnd.AppendLine("jpeg;");
cmnd.AppendLine("scale 1;");
cmnd.AppendLine(" psize 8;");
cmnd.AppendLine("nojitter;");
cmnd.AppendLine("full;");
cmnd.AppendLine("bound;");
cmnd.AppendLine("noal;");
cmnd.AppendLine("regr;");
cmnd.AppendLine("symb;");
cmnd.AppendLine("nodt.");
//Calculate correlation coefficient
string[] matIds = Mtblib.Tools.MtbTools.CreateVariableStrArray(ws, 1, Mtblib.Tools.MtbVarType.Matrix);
cmnd.AppendFormat("corr {0} {1} {2}\r\n", string.Join(" &\r\n", bkCols.Select(x => x.SynthesizedName)),
string.Join(" &\r\n", varCols.Select(x => x.SynthesizedName)),
string.Join(" &\r\n", matIds)
);
//刪除所有圖形
cmnd.AppendLine("gmana;");
cmnd.AppendLine("all;");
cmnd.AppendLine("close;");
cmnd.AppendLine("nopr.");
string fpath = Mtblib.Tools.MtbTools.BuildTemporaryMacro("mymacro.mtb", cmnd.ToString());
proj.ExecuteCommand(string.Format("exec \"{0}\" 1", fpath), ws);
//將檔案轉為二進位陣列
string _root = System.IO.Path.Combine(Environment.GetEnvironmentVariable("tmp"), "Minitab", string.Format("trend_{0}", _rawdata.TableName));
string _file;
for (int i = 0; i < bkCols.Count + varCols.Count; i++)
{
_file = _root + (i + 1).ToString("D3") + ".jpg";
if (File.Exists(_file))
{
this.Contents.Add(new RptOutput()
{
OType = MtbOType.GRAPH,
OutputInByteArr = File.ReadAllBytes(_file),
Tag = "Trend"
});
}
}
//if (File.Exists(gpath_trend))
//{
// this.Contents.Add(new RptOutput()
// {
// OType = MtbOType.GRAPH,
// OutputInByteArr = File.ReadAllBytes(gpath_trend),
// Tag = "Trend"
// });
//}
if (File.Exists(gpath_corr))
{
this.Contents.Add(new RptOutput()
{
OType = MtbOType.GRAPH,
OutputInByteArr = File.ReadAllBytes(gpath_corr),
Tag = "Scatter"
});
}
//Create Correlation table
DataTable corrTable = new DataTable();
corrTable.Columns.Add("熔爐項目", typeof(string));
foreach (Mtb.Column col in bkCols)
{
corrTable.Columns.Add(col.Name, typeof(double));
}
//corrTable.Columns.Add("CorrCoef", typeof(double));
var corrMat = ws.Matrices.Item(matIds[0]).GetData(); //取出 correlation matrix
var mBuilder = LinearAlgebra.Matrix <double> .Build;
LinearAlgebra.Matrix <double> corM = mBuilder.DenseOfColumnMajor(bkCols.Count + varCols.Count, bkCols.Count + varCols.Count, corrMat);
LinearAlgebra.Matrix <double> subCorrMat = corM.SubMatrix(bkCols.Count, varCols.Count, 0, bkCols.Count);
string[] labNames = varCols.Select(x => x.Name).ToArray();
List <object> itemArray;
DataRow dr;
for (int i = 0; i < varCols.Count; i++)
{
itemArray = new List <object>();
itemArray.Add(labNames[i]);
itemArray.AddRange(subCorrMat.Row(i).ToArray().Select(x => (object)Math.Round(x, 3)).ToArray());
//dr = corrTable.NewRow();
//dr.ItemArray = itemArray;
corrTable.Rows.Add(itemArray.ToArray());
}
//double[] subCorr = new double[varCols.Count];
//double[] corr = corrMat as double[];
//if (corr != null)
//{
// Array.Copy(corr, 1, subCorr, 0, varCols.Count);
//
// DataRow dr;
// for (int i = 0; i < subCorr.Length; i++)
// {
// dr = corrTable.NewRow();
// dr[0] = labNames[i];
// dr[1] = subCorr[i];
// corrTable.Rows.Add(dr);
// }
//}
this.Contents.Add(new RptOutput()
{
OType = MtbOType.TABLE,
OutputInByteArr = Tool.ConvertDataSetToByteArray(corrTable),
Tag = "Table"
});
}