Soviet laboratory with sensors on concrete walls

Today I am reminded of one of my most ambitious (and most ridiculed) research projects from my time at Laboratory 23-Б: the seven-year study of temperature fluctuations in concrete structures.

The Genesis

It began in 1998 when I noticed unusual temperature readings on a thermocouple embedded in the concrete wall of our laboratory. The variations were small - barely 0.3°C - but they followed a pattern that seemed independent of external temperature, humidity, or any obvious environmental factors.

My supervisor, Dr. Yevgeny Konstantinovich, suggested it was “probably nothing.” I decided to investigate further.

Methodology

Over the next seven years, I:

  1. Installed 47 thermocouples in various concrete structures around the facility
  2. Recorded temperature readings every 15 minutes
  3. Collected approximately 1.3 million data points
  4. Cross-referenced with lunar phases, solar activity, local seismic data, and barometric pressure
  5. Analyzed the data using every statistical method I could find (and several I invented)

Findings

After exhaustive analysis, I discovered:

  • The temperature variations were real and reproducible
  • They correlated weakly (r=0.23) with lunar phase, but only in concrete poured on Tuesdays
  • Concrete containing aggregate from the Kokshetau quarry showed different patterns than concrete from other sources
  • The effect disappeared entirely in concrete structures built after 2001

Publication

I submitted my findings to 12 journals. Here is a summary of responses:

  • 8 journals: Desk rejection
  • 2 journals: Sent for review, then rejected with comments like “interesting but inconclusive”
  • 1 journal: Accepted with major revisions, then retracted when I couldn’t replicate the Tuesday effect
  • 1 journal: Published! (The Proceedings of the Almaty Physical Society, which ceased publication the following year)

Current Theory

After many years of reflection, I believe the temperature variations were caused by:

  1. Gradual curing processes in the concrete (30% probability)
  2. Unknown environmental factors I failed to measure (40% probability)
  3. Measurement artifacts from aging thermocouples (25% probability)
  4. Genuine physical phenomena that I don’t understand (5% probability)

My colleague Sergei suggested a fifth option: “You are measuring random noise and seeing patterns because humans are pattern-seeking creatures.” I cannot completely rule this out.

Lessons Learned

  • Seven years is a long time to spend measuring concrete
  • Statistical significance does not always imply physical significance
  • Always calibrate your thermocouples (I learned this in year 4)
  • Academic reviewers are not impressed by sample sizes alone
  • Sometimes “probably nothing” is actually nothing

Current Status

I still have all 1.3 million data points on a hard drive somewhere. Occasionally I think about re-analyzing them with modern machine learning techniques. Then I remember that this would require learning modern machine learning techniques, and I go do something else.

Would I do it again? Absolutely not. Am I glad I did it? Ask me in another seven years.

Note: If anyone reading this has experience with thermal properties of concrete and can explain why Tuesday-poured concrete behaves differently, please contact me. I am still curious about this.