How to repair a corrupt mat file?

Hello,

we have made measurements with a Bachmann controller and Simulink. More precisely, we used a Simulink block provided by Bachmann that allows to start and stop saving measurements to a .mat file. While saving, the .mat file located on the Bachmann memory "grows".

However, one time we did not stop the saving properly, but the connection to the sensors was interrupted and the saving was simply cancelled. Now we cannot load the .mat file into the Matlab workspace. The error message is "File might be corrupt".

Here we could not upload the .mat file because it is too big (268MB).

We would appreciate any advice on how to fix the corrupt .mat file. If necessary, we can of course try to make the .mat file available by other means.

Thank you very much!

Amelie

[dpb]

If using the default -v73, they are HDF5 under the hood and h5info and friends may be of some help.

If there are multiple variables in the .mat file, sometimes you can load specific variables from the .mat file with the load('filename.mat','variablename') syntax, ignoring other/corrupted individual variables. If it is all one large array, this obviously doesn't help.

Beyond that, you're basically out of luck.

While convenient syntactically, it seems that .mat files tend to be fragile although a failure to close properly can corrupt any file, not just .mat files. But, I've moved away from using them for really valuable data in favor of regular stream files. This has the downside of having to have specific code to read/write them, but I've not had them get corrupted the same way as .mat files can be.

[Walter Roberson]

If it is a -v7.0 file (most common), then https://www.mathworks.com/matlabcentral/fileexchange/42274-loadfixnames-loads-a-mat-file-fixing-invalid-names might be able to give information about the file and possibly extract variables.

[Amelie Zeller]

I tried all the suggested functions.

Unfortunately, we stored the data in a single variable. The h5info function didn't work for some reason and the loadfixnames function also didn't work.

I tried to open the .mat file in a text editor to manually fix it, but also this didn't work...

[Walter Roberson]

.mat files are binary files, and are not expected to make sense in a text editor (other than roughly the first 40 bytes)

[dpb]

"I tried to open the .mat file in a text editor..."

No, unless the -text flag is used, .mat files are an unformatted file; the only way to read them other than by TMW-supplied load or one of the FEX tools would be to use the file structure and read the raw data.

The first N bytes in a .mat file are text header that will identify the version, assuming the whole file has not been totally trashed somehow. try

type yourmatfile.mat

and see what it looks like -- first few lines followed by a bunch of gibberish. If it's large you might just read the first few lines and echo to the screen, or you should have been able to see the header in the text editor to identify at least the version.

There are routines in the external language interface section to read/write .mat files, whether they can be used on a corrupt file or whether they'll also barf on input checking first I don't know...

https://www.mathworks.com/help/pdf_doc/matlab/matfile_format.pdf is a description of the version 4 (unlikely) and 5 file formats; for later versions I don't know that it is available.

[Bruno Luong]

IMO, the best chance is soliciting TMW consulting service to eventually recover your mat file.