The results (logFC) are not consistence (edgeR packages)

I'm trying to find the differences between the test and control groups (genes). But the results are not consistence. For example, when there's a decrease in the test group, the fold change is positive! This happens in about 5/200 of them. Here's my codes, I don't know what's wrong... (because of the signed contact, I can't post any data here, sorry!) :

The data (df) is like:

gene Test1 Test2 Test3 Control1 Control2 Control3 Control4

A ...

B ...

C ...
...
n = 200

groups = factor(c(rep("test",3), rep("control",4)))
dge = DGEList(as.matrix(df), group = groups) 
dge = calcNormFactors(dge, method = "TMM") 
design_matrix = model.matrix(~0+groups, data = dge$samples) 
colnames(design_matrix) = c("test", "control") 
dge = estimateDisp(dge, design_matrix, robust=TRUE) 
FIT = glmQLFit(dge, design_matrix, robust=TRUE)
model = glmQLFTest(FIT, contrast = makeContrasts(contrasts = c("test-control"), levels = design_matrix)) 
output = data.frame(Gene = rownames(dge$counts), FoldChange = 2^-(model$table$logFC))

thanks in advance!